WEBVTT
1
00:00:00.000 --> 00:00:00.833
Yes.
2
00:00:02.150 --> 00:00:04.790
I wanna welcome everyone to day two
3
00:00:04.790 --> 00:00:08.620
of the NIDA-NIAAA Mini-Convention.
4
00:00:08.620 --> 00:00:11.730
My name is John Matochik, I'm a Program Officer
5
00:00:11.730 --> 00:00:14.760
in the Division of Neuroscience and Behavior
6
00:00:14.760 --> 00:00:18.610
at the National Institute on Alcohol Abuse and Alcoholism,
7
00:00:18.610 --> 00:00:21.380
and along with my colleague, Roger Sorenson,
8
00:00:21.380 --> 00:00:25.890
who you met yesterday, we are the co-chairs
9
00:00:25.890 --> 00:00:30.890
for the NIDA-NIAAA Neuroscience Working Group,
10
00:00:30.957 --> 00:00:33.040
and our primary responsibility
11
00:00:33.040 --> 00:00:35.970
is putting on the mini-convention each year.
12
00:00:35.970 --> 00:00:38.220
And we're really pleased to do it this year.
13
00:00:38.220 --> 00:00:41.100
And I think we've gotten a lot of positive comments
14
00:00:41.100 --> 00:00:43.270
about yesterday and it was just really good.
15
00:00:43.270 --> 00:00:46.590
And I think today's gonna be even just as good.
16
00:00:46.590 --> 00:00:49.320
And I'm really excited because we're gonna start
17
00:00:49.320 --> 00:00:52.420
the Early Career Investigators Showcase.
18
00:00:52.420 --> 00:00:54.670
This over the last couple of years has become
19
00:00:54.670 --> 00:00:58.713
a very popular event at the mini-convention.
20
00:00:59.870 --> 00:01:04.870
This year both NIAAA and NIDA able to give four awards each.
21
00:01:06.360 --> 00:01:09.370
So we'll have eight presenters today.
22
00:01:09.370 --> 00:01:12.620
These are all early career stage investigators
23
00:01:12.620 --> 00:01:16.420
presenting their research in very brief talks.
24
00:01:16.420 --> 00:01:19.110
And the talks will be about seven minutes long
25
00:01:19.110 --> 00:01:22.250
followed by about three minutes for questions.
26
00:01:22.250 --> 00:01:25.590
So we might only be able to get one or two per talk.
27
00:01:25.590 --> 00:01:28.030
And then we have a little bit of time at the end
28
00:01:28.030 --> 00:01:32.920
we hope that any unanswered questions, we can address.
29
00:01:32.920 --> 00:01:36.060
I also wanna remind the audience that the posters
30
00:01:36.060 --> 00:01:39.820
with more detail about their research is online
31
00:01:39.820 --> 00:01:42.520
and this will stay online for a number of weeks.
32
00:01:42.520 --> 00:01:46.180
So if you're interested, certainly go see their poster.
33
00:01:46.180 --> 00:01:47.530
And I really encourage you
34
00:01:47.530 --> 00:01:50.640
to also contact these young investigators.
35
00:01:50.640 --> 00:01:53.770
This is the future of addiction research.
36
00:01:53.770 --> 00:01:56.080
And I have to say, we have been very pleased
37
00:01:56.080 --> 00:02:00.530
at both NIDA and NIAAA with the applications we received.
38
00:02:00.530 --> 00:02:03.180
We received far too many good applications
39
00:02:03.180 --> 00:02:06.640
it makes it very difficult to pick our selections
40
00:02:06.640 --> 00:02:10.010
for the mini-convention, which to us is a very good sign
41
00:02:10.010 --> 00:02:13.900
about the vitality of the new generation coming up.
42
00:02:13.900 --> 00:02:16.190
This one of know, one of the other things
43
00:02:16.190 --> 00:02:21.190
that the NIDA and NIAAA Neuroscience Working Group does
44
00:02:22.050 --> 00:02:25.780
is every month we have a monthly meeting of program officers
45
00:02:25.780 --> 00:02:28.890
with both institutes to talk about various issues.
46
00:02:28.890 --> 00:02:30.680
Obviously the big issue is planning
47
00:02:30.680 --> 00:02:32.610
each year's mini-convention,
48
00:02:32.610 --> 00:02:34.270
but also in each month's meeting
49
00:02:34.270 --> 00:02:36.390
we kinda have a journal club.
50
00:02:36.390 --> 00:02:38.950
And instead of us presenting articles,
51
00:02:38.950 --> 00:02:41.400
we started about two, three years ago
52
00:02:41.400 --> 00:02:44.200
having young and junior investigators present
53
00:02:44.200 --> 00:02:47.449
to us, and what this does, it allows them
54
00:02:47.449 --> 00:02:50.410
to really promote themselves in a research
55
00:02:50.410 --> 00:02:53.660
and it helps us as program officers to know
56
00:02:53.660 --> 00:02:55.660
who are the up-and-coming people
57
00:02:55.660 --> 00:02:58.200
and what support we can give them.
58
00:02:58.200 --> 00:03:00.180
And it's also a way program officers
59
00:03:00.180 --> 00:03:04.000
to show off some of the young people in their portfolios.
60
00:03:04.000 --> 00:03:06.923
And the eight people speaking today,
61
00:03:06.923 --> 00:03:09.997
maybe down the road, one of them
62
00:03:09.997 --> 00:03:13.180
will be given the Jacob Alec's Award,
63
00:03:13.180 --> 00:03:15.840
I hope you all didn't miss that excellent lecture
64
00:03:15.840 --> 00:03:18.353
by Lorenzo yesterday.
65
00:03:18.353 --> 00:03:19.740
So again, we're gonna hear
66
00:03:19.740 --> 00:03:22.930
from the next generation of addiction researchers.
67
00:03:22.930 --> 00:03:25.870
So without further ado, let's get started.
68
00:03:25.870 --> 00:03:27.350
And I'd like to call
69
00:03:27.350 --> 00:03:30.760
on Alex Ramsey from Washington University
70
00:03:30.760 --> 00:03:33.263
in St. Louis to give the first presentation.
71
00:03:35.690 --> 00:03:36.750
Thank you so much.
72
00:03:36.750 --> 00:03:39.330
It's really a (clears throat) privilege
73
00:03:39.330 --> 00:03:41.270
to be here with you this morning.
74
00:03:41.270 --> 00:03:43.390
My name is Alex Ramsey,
75
00:03:43.390 --> 00:03:47.050
Assistant Professor of Psychiatry at Washington University.
76
00:03:47.050 --> 00:03:48.540
And I'm really coming at this work
77
00:03:48.540 --> 00:03:52.930
from the other end of the translational science paradigm.
78
00:03:52.930 --> 00:03:55.160
My work is in implementation science related
79
00:03:55.160 --> 00:03:58.140
to substance use disorder, and I'm funded
80
00:03:58.140 --> 00:04:00.680
on NIDA K12 to study that process
81
00:04:00.680 --> 00:04:03.920
of moving genomic applications
82
00:04:03.920 --> 00:04:08.920
from basic genomic discovery and toward implementation
83
00:04:10.070 --> 00:04:11.940
in clinical and community settings.
84
00:04:11.940 --> 00:04:14.770
So it's a pleasure to be presenting this work
85
00:04:14.770 --> 00:04:16.794
on advancing the development
86
00:04:16.794 --> 00:04:20.220
of a genetically-informed smoking cessation intervention.
87
00:04:20.220 --> 00:04:22.506
And on the first slide, we'll see
88
00:04:22.506 --> 00:04:24.480
that this is really rooted
89
00:04:24.480 --> 00:04:27.160
in the translational science paradigm
90
00:04:27.160 --> 00:04:28.880
as we move from genomic discovery
91
00:04:28.880 --> 00:04:31.520
to genetically-informed behavioral interventions.
92
00:04:31.520 --> 00:04:36.030
And we really leveraged the NIH Stage Model
93
00:04:36.030 --> 00:04:37.483
as this rolls out,
94
00:04:38.460 --> 00:04:41.690
which really begins with stage zero, basic research.
95
00:04:41.690 --> 00:04:43.900
And you'll see things pop up showing
96
00:04:43.900 --> 00:04:48.700
how we characterize this, with genomic discovery
97
00:04:48.700 --> 00:04:52.240
being our basic science research at stage zero
98
00:04:52.240 --> 00:04:53.690
and then moving into stage one
99
00:04:53.690 --> 00:04:58.690
where we iteratively develop a behavioral intervention tool
100
00:04:58.840 --> 00:05:00.580
and then conducted feasibility testing
101
00:05:00.580 --> 00:05:02.360
and proof-of-concept testing.
102
00:05:02.360 --> 00:05:05.370
This work then moves into efficacy and effectiveness testing
103
00:05:05.370 --> 00:05:07.610
in real-world settings, and then ultimately
104
00:05:07.610 --> 00:05:11.230
into dissemination and implementation research.
105
00:05:11.230 --> 00:05:12.130
Next slide please.
106
00:05:14.860 --> 00:05:18.430
And then, so on the next slide, you'll see that,
107
00:05:18.430 --> 00:05:21.700
so I'll be telling a short story on the genetics
108
00:05:21.700 --> 00:05:25.140
of smoking as we bridge the past, present and future.
109
00:05:25.140 --> 00:05:28.770
I refer to genomic discovery as the past, although this
110
00:05:28.770 --> 00:05:31.920
is of course, a very active area of investigation,
111
00:05:31.920 --> 00:05:35.935
the president being our work to develop and conduct
112
00:05:35.935 --> 00:05:36.820
proof-of-concept testing
113
00:05:36.820 --> 00:05:38.830
for this behavioral intervention tool.
114
00:05:38.830 --> 00:05:42.860
And then the future really reflects our work to move this
115
00:05:42.860 --> 00:05:44.220
into real-world context
116
00:05:44.220 --> 00:05:49.220
and investigate clinical utility of these tools.
117
00:05:49.390 --> 00:05:53.800
So on the next slide, we should see
118
00:05:53.800 --> 00:05:57.350
that we now have evidence that variants
119
00:05:57.350 --> 00:06:00.470
in and near the CHRNA5 gene region
120
00:06:00.470 --> 00:06:02.800
have prognostic significance
121
00:06:02.800 --> 00:06:05.100
for one's risk of smoking-related diseases,
122
00:06:05.100 --> 00:06:07.870
their likelihood to be able to quit smoking
123
00:06:07.870 --> 00:06:10.020
and their response to treatment.
124
00:06:10.020 --> 00:06:14.320
And we know that these individuals, should pop up next,
125
00:06:14.320 --> 00:06:18.180
that that have high-risk genetic variants tend
126
00:06:18.180 --> 00:06:22.800
to smoke more heavily, have increased risk for lung cancer,
127
00:06:22.800 --> 00:06:27.020
develop lung cancer sooner, quit smoking at a later age
128
00:06:27.020 --> 00:06:31.210
and ultimately have less success with quit attempts.
129
00:06:31.210 --> 00:06:35.000
So we then took this genomic discovery data
130
00:06:35.000 --> 00:06:38.430
and began to iteratively design and develop the tool.
131
00:06:38.430 --> 00:06:40.310
So on the next slide,
132
00:06:40.310 --> 00:06:45.310
we know that demonstrating high demand
133
00:06:45.500 --> 00:06:49.860
for a behavioral intervention tool is absolutely critical
134
00:06:49.860 --> 00:06:51.830
to the implementation of that tool.
135
00:06:51.830 --> 00:06:53.040
So we were very pleased to see
136
00:06:53.040 --> 00:06:55.250
that within our current smoker population,
137
00:06:55.250 --> 00:06:58.290
there was very robust and high levels of interest
138
00:06:58.290 --> 00:07:01.970
among this group, with a 95% indicating
139
00:07:01.970 --> 00:07:04.590
that they wanted this personalized genetic information
140
00:07:04.590 --> 00:07:06.353
to guide their smoking cessation.
141
00:07:07.400 --> 00:07:12.320
Next up, we then iteratively designed a tool
142
00:07:12.320 --> 00:07:16.190
in coordination and collaboration with current smokers
143
00:07:16.190 --> 00:07:19.770
as well as behavioral health counselors.
144
00:07:19.770 --> 00:07:21.070
And I just wanna give you a sense
145
00:07:21.070 --> 00:07:22.300
of what this tool looks like,
146
00:07:22.300 --> 00:07:24.730
and you'll see it sort of roll out here.
147
00:07:24.730 --> 00:07:29.730
We conducted a brief orientation of genetics
148
00:07:30.080 --> 00:07:32.830
and where we were looking along
149
00:07:32.830 --> 00:07:35.860
the genome for the individual,
150
00:07:35.860 --> 00:07:40.270
and then as this rolls out, go ahead and roll this out.
151
00:07:40.270 --> 00:07:43.490
We basically have, we returned
152
00:07:43.490 --> 00:07:45.220
personalized genetic information
153
00:07:45.220 --> 00:07:48.690
to the individual, categorized their risk,
154
00:07:48.690 --> 00:07:52.800
interpreted that risk and then ultimately framed that risk
155
00:07:52.800 --> 00:07:56.990
in terms of behavioral intervention messaging
156
00:07:56.990 --> 00:07:59.310
and risk communication messaging,
157
00:07:59.310 --> 00:08:01.620
and ultimately provided quick advice
158
00:08:01.620 --> 00:08:06.620
and easy-to-access resources to help that individual quit.
159
00:08:06.910 --> 00:08:08.063
Next slide, please.
160
00:08:09.586 --> 00:08:14.270
You'll see that we use a four-step process
161
00:08:14.270 --> 00:08:18.050
to categorize risk on this using this risk profile tool.
162
00:08:18.050 --> 00:08:20.843
And it began with selecting CHRNA5
163
00:08:20.843 --> 00:08:24.350
that were selected by ancestry.
164
00:08:24.350 --> 00:08:27.700
We then combined this information
165
00:08:27.700 --> 00:08:32.700
with the heaviness of smoking for the individual.
166
00:08:33.810 --> 00:08:35.560
And so we use both the genetics
167
00:08:35.560 --> 00:08:39.950
and the smoking heaviness to combine that
168
00:08:39.950 --> 00:08:43.600
into three categories of risk.
169
00:08:43.600 --> 00:08:47.110
And then on the final slide, moving forward,
170
00:08:47.110 --> 00:08:49.510
we then place that individual along a spectrum
171
00:08:49.510 --> 00:08:54.370
of risk, and for three smoking-related phenotypes
172
00:08:54.370 --> 00:08:55.970
of lung cancer, lung disease
173
00:08:55.970 --> 00:08:57.560
and difficulty quitting smoking.
174
00:08:57.560 --> 00:08:58.510
Next slide, please.
175
00:08:59.880 --> 00:09:03.550
Then finally, I'll discuss our proof-of-concept testing.
176
00:09:03.550 --> 00:09:04.693
Next, please.
177
00:09:05.580 --> 00:09:08.130
And this is, you can go ahead and roll this out.
178
00:09:08.130 --> 00:09:12.170
This is a, we recruited 111 current smokers
179
00:09:12.170 --> 00:09:15.000
for baseline assessment and genetic testing.
180
00:09:15.000 --> 00:09:18.781
We're pleased to see that 108
181
00:09:18.781 --> 00:09:20.697
of those had successfully processed genetic results
182
00:09:20.697 --> 00:09:25.697
and were invited to visit two where we returned
183
00:09:26.380 --> 00:09:28.450
the risk profile intervention,
184
00:09:28.450 --> 00:09:31.740
and then visit three was our 30-day followup assessment.
185
00:09:31.740 --> 00:09:32.780
And we're very happy
186
00:09:32.780 --> 00:09:35.900
with high levels of retention across all visits.
187
00:09:35.900 --> 00:09:37.163
Next slide, please.
188
00:09:38.140 --> 00:09:41.290
The risk profile was highly acceptable.
189
00:09:41.290 --> 00:09:43.557
There was very little decisional regret,
190
00:09:43.557 --> 00:09:46.920
high levels of comprehension and recall
191
00:09:47.820 --> 00:09:51.800
at the 30-day follow-up, and over 90% found the tool
192
00:09:51.800 --> 00:09:55.150
to be useful to extremely useful.
193
00:09:55.150 --> 00:09:58.360
And so on the next slide, we can see the reduction
194
00:09:58.360 --> 00:10:00.320
in smoking over time as participants moved
195
00:10:00.320 --> 00:10:02.840
from about 13 cigarettes per day
196
00:10:02.840 --> 00:10:04.730
to less than 10 cigarettes per day,
197
00:10:04.730 --> 00:10:07.070
including a significant decrease
198
00:10:07.070 --> 00:10:10.960
in smoking at 30-day post-intervention.
199
00:10:10.960 --> 00:10:11.963
Next slide.
200
00:10:13.990 --> 00:10:18.990
So we can see that about 70% of individuals
201
00:10:20.120 --> 00:10:22.680
became more ready to quit or do smoking,
202
00:10:22.680 --> 00:10:26.430
about two thirds indicated reducing their level
203
00:10:26.430 --> 00:10:29.810
of smoking, and a large proportion
204
00:10:29.810 --> 00:10:31.890
indicated they had made a quit attempt,
205
00:10:31.890 --> 00:10:35.045
had begun to use smoking cessation medications,
206
00:10:35.045 --> 00:10:38.670
and about 10% reported having quit.
207
00:10:38.670 --> 00:10:41.070
And importantly, there were no clear differences
208
00:10:41.070 --> 00:10:43.350
by risk level, indicating
209
00:10:45.233 --> 00:10:48.210
that there was robust support across levels of risks.
210
00:10:48.210 --> 00:10:49.043
Next slide.
211
00:10:50.780 --> 00:10:53.403
So in conclusion, we leveraged genomic discovery data
212
00:10:53.403 --> 00:10:58.315
and the NIH Stage Model, found very high levels
213
00:10:58.315 --> 00:11:00.270
of acceptability and proof-of-concept data
214
00:11:00.270 --> 00:11:01.860
to support moving forward.
215
00:11:01.860 --> 00:11:02.750
And as we look ahead,
216
00:11:02.750 --> 00:11:06.310
we'll be designing this tool for implementation,
217
00:11:06.310 --> 00:11:08.410
continuing to think about the people, tools
218
00:11:08.410 --> 00:11:10.700
and systems needed for implementation
219
00:11:10.700 --> 00:11:15.470
as we seek out areas of high demand for this tool
220
00:11:15.470 --> 00:11:18.660
in primary care settings, lung cancer screening,
221
00:11:18.660 --> 00:11:21.350
working in behavioral health care settings.
222
00:11:21.350 --> 00:11:23.010
So that's absolutely critical
223
00:11:23.010 --> 00:11:25.590
for implementation work moving forward.
224
00:11:25.590 --> 00:11:28.610
So just want to acknowledge the support
225
00:11:28.610 --> 00:11:30.710
of many people as well as funding
226
00:11:31.649 --> 00:11:34.549
from NIDA, and want to thank everyone for their attention.
227
00:11:37.890 --> 00:11:38.723
Okay.
228
00:11:38.723 --> 00:11:42.610
We have, I think like one question in the chat.
229
00:11:42.610 --> 00:11:44.230
Very nice talk.
230
00:11:44.230 --> 00:11:46.780
Is anyone pursuing such an approach
231
00:11:46.780 --> 00:11:49.530
with the low, low response alcohol group
232
00:11:49.530 --> 00:11:53.127
of individuals with a family history of AUD?
233
00:11:56.280 --> 00:11:57.113
Great question.
234
00:11:57.113 --> 00:11:59.400
So that is an area (clears throat)
235
00:11:59.400 --> 00:12:04.170
We still see that as an unexplored area of research.
236
00:12:04.170 --> 00:12:09.100
I'll say that for this work, we began in the area of smoking
237
00:12:09.100 --> 00:12:12.437
because we felt like the data were most strong
238
00:12:12.437 --> 00:12:17.437
and this was a ripe area to begin to classify
239
00:12:18.110 --> 00:12:22.080
this risk information and begin returning this information
240
00:12:22.080 --> 00:12:26.530
to see if we can detect behavior change
241
00:12:26.530 --> 00:12:29.550
and begin to see, is this the type
242
00:12:29.550 --> 00:12:33.150
of information at first that people want to receive,
243
00:12:33.150 --> 00:12:36.150
and can it get people to begin thinking differently
244
00:12:36.150 --> 00:12:37.000
about a quit attempt?
245
00:12:37.000 --> 00:12:39.410
But I think these are successes
246
00:12:39.410 --> 00:12:41.380
that we're beginning to see that can absolutely begin
247
00:12:41.380 --> 00:12:45.920
to be explored with other substance uses
248
00:12:45.920 --> 00:12:50.920
and notably within individuals with risky alcohol use
249
00:12:51.970 --> 00:12:54.963
and alcohol use disorder as well, so.
250
00:12:54.963 --> 00:12:55.796
Okay, we have one other question,
251
00:12:55.796 --> 00:12:57.930
I think we can fit in one other question.
252
00:12:57.930 --> 00:13:01.810
The question is, what treatments drugs were used
253
00:13:01.810 --> 00:13:05.053
for smoking cessations for the different groups?
254
00:13:05.920 --> 00:13:09.170
So this was a, we conducted
255
00:13:09.170 --> 00:13:10.970
this proof-of-concept study naturally.
256
00:13:10.970 --> 00:13:15.970
We did not actually provide access to treatments.
257
00:13:18.720 --> 00:13:21.010
We made strong recommendations
258
00:13:21.010 --> 00:13:24.470
based on the individuals as part
259
00:13:24.470 --> 00:13:27.010
of the behavioral intervention and provided support
260
00:13:27.010 --> 00:13:32.010
for how they could tap into QuitLine resources,
261
00:13:33.160 --> 00:13:38.060
text and app-based resources and discuss
262
00:13:38.060 --> 00:13:41.690
with them how they could access nicotine replacement therapy
263
00:13:41.690 --> 00:13:45.680
or (indistinct) clean in consultation
264
00:13:45.680 --> 00:13:47.957
with their primary care physician.
265
00:13:47.957 --> 00:13:50.680
But providing actual access
266
00:13:50.680 --> 00:13:54.600
to the drugs was not part of the work to date,
267
00:13:54.600 --> 00:13:59.510
although that is our plan setting moving forward.
268
00:13:59.510 --> 00:14:00.960
Okay, thank you Alex.
269
00:14:00.960 --> 00:14:04.260
We're gonna move on to the next Early Career Investigator,
270
00:14:04.260 --> 00:14:08.600
and that's Emma Johnson, also from the Washington University
271
00:14:08.600 --> 00:14:09.910
in St. Louis,
272
00:14:09.910 --> 00:14:11.203
Emma, it's all yours.
273
00:14:14.770 --> 00:14:16.520
Thank you so much.
274
00:14:16.520 --> 00:14:19.880
Today I'm really excited to talk to you all about our work
275
00:14:19.880 --> 00:14:22.530
looking at the relationships between cannabis, tobacco
276
00:14:22.530 --> 00:14:24.570
and schizophrenia, and how we can explore these
277
00:14:24.570 --> 00:14:27.050
from a genome-wide perspective.
278
00:14:27.050 --> 00:14:28.193
Next slide, please.
279
00:14:29.140 --> 00:14:31.060
Schizophrenia is a very serious
280
00:14:31.060 --> 00:14:33.670
and debilitating disorder, and it's well-known
281
00:14:33.670 --> 00:14:35.870
that problematic substance use often co-occurs
282
00:14:35.870 --> 00:14:39.020
with schizophrenia in particular, the relationship
283
00:14:39.020 --> 00:14:42.150
between cannabis use and schizophrenia has been one
284
00:14:42.150 --> 00:14:45.930
of enduring interest in the psychiatry field.
285
00:14:45.930 --> 00:14:47.500
And there are a couple of common theories
286
00:14:47.500 --> 00:14:49.600
for why we see this co-morbidity.
287
00:14:49.600 --> 00:14:51.920
The first is that there's actually a causal relationship
288
00:14:51.920 --> 00:14:54.540
whereby cannabis use causes the development
289
00:14:54.540 --> 00:14:56.170
of psychosis and schizophrenia.
290
00:14:56.170 --> 00:14:58.370
The second theory
291
00:14:58.370 --> 00:15:01.070
is that they're actually shared genetic influences.
292
00:15:01.070 --> 00:15:03.160
So genetic factors that increase risk
293
00:15:03.160 --> 00:15:04.926
of cannabis use may also increase risk
294
00:15:04.926 --> 00:15:07.177
of developing schizophrenia.
295
00:15:07.177 --> 00:15:09.550
And both of these mechanisms likely play a role
296
00:15:09.550 --> 00:15:11.380
in this co-morbidity, but I'm really gonna focus
297
00:15:11.380 --> 00:15:12.893
on that second point today.
298
00:15:14.239 --> 00:15:15.072
So there've been lots of studies
299
00:15:15.072 --> 00:15:17.090
over the years that have tried to resolve these questions,
300
00:15:17.090 --> 00:15:19.560
but I think a few gaps remain.
301
00:15:19.560 --> 00:15:21.480
The first is that there's increasing evidence
302
00:15:21.480 --> 00:15:24.152
that genetic factors underlying cannabis every-use,
303
00:15:24.152 --> 00:15:26.266
so whether you've ever tried cannabis in your lifetime
304
00:15:26.266 --> 00:15:29.010
may at least partially diverge
305
00:15:29.010 --> 00:15:31.850
from the genetic factors that underlie more severe measures
306
00:15:31.850 --> 00:15:34.950
of use like cannabis use disorder.
307
00:15:34.950 --> 00:15:36.420
And the studies that have looked
308
00:15:36.420 --> 00:15:38.660
at these questions have typically not looked
309
00:15:38.660 --> 00:15:40.603
at these more severe phenotypes.
310
00:15:41.750 --> 00:15:44.980
We also know that tobacco is correlated both phenotypically
311
00:15:44.980 --> 00:15:48.030
and genetically with cannabis use and schizophrenia.
312
00:15:48.030 --> 00:15:49.380
So this makes it a potential confounder.
313
00:15:49.380 --> 00:15:51.130
And it's really important
314
00:15:51.130 --> 00:15:54.300
that we include this in our future analysis.
315
00:15:54.300 --> 00:15:56.110
Finally, to my knowledge,
316
00:15:56.110 --> 00:15:57.300
there's been no study yet
317
00:15:57.300 --> 00:15:58.850
that has systematically examined
318
00:15:58.850 --> 00:16:00.530
the specific genetic variance
319
00:16:00.530 --> 00:16:02.330
in genes that may contribute
320
00:16:02.330 --> 00:16:06.070
to both problematic cannabis use and schizophrenia.
321
00:16:06.070 --> 00:16:07.153
Next slide, please.
322
00:16:09.040 --> 00:16:11.840
So these are the genome-wide association study datasets
323
00:16:11.840 --> 00:16:14.760
that I used in the analysis that I'll present to you today.
324
00:16:14.760 --> 00:16:18.710
So first we used GWAS of schizophrenia from the PGC.
325
00:16:18.710 --> 00:16:21.960
For cannabis we included a GWAS of lifetime ever-use
326
00:16:21.960 --> 00:16:25.210
and a GWAS analysis of cannabis use disorder.
327
00:16:25.210 --> 00:16:27.010
And for tobacco, we looked at a GWAS
328
00:16:27.010 --> 00:16:30.380
of ever smoking and a GWAS of nicotine dependence
329
00:16:30.380 --> 00:16:31.853
that was recently published.
330
00:16:32.980 --> 00:16:34.023
Next slide please.
331
00:16:35.510 --> 00:16:37.980
So the analysis I present to you today broadly fall
332
00:16:37.980 --> 00:16:39.730
into two categories.
333
00:16:39.730 --> 00:16:42.050
First, I was really interested in looking
334
00:16:42.050 --> 00:16:44.790
at the associations between cannabis, tobacco
335
00:16:44.790 --> 00:16:46.330
and schizophrenia for more
336
00:16:46.330 --> 00:16:48.520
of a broad trait-level perspective.
337
00:16:48.520 --> 00:16:51.493
And to do this, we use genomic structural equation models.
338
00:16:52.370 --> 00:16:54.600
Second, we were interested in delving a little bit
339
00:16:54.600 --> 00:16:57.280
more deeply and looking at the specific genetic variants
340
00:16:57.280 --> 00:16:58.960
in genes that contribute
341
00:16:58.960 --> 00:17:01.390
to develop cannabis use disorder and schizophrenia.
342
00:17:01.390 --> 00:17:03.200
And to do this, I used something called ASSET,
343
00:17:03.200 --> 00:17:06.523
which is a subset-based association approach.
344
00:17:07.370 --> 00:17:09.840
So on the next slide, I'm presenting
345
00:17:09.840 --> 00:17:12.990
our first genomic structural equation model here.
346
00:17:12.990 --> 00:17:14.350
And you can think of this
347
00:17:14.350 --> 00:17:16.970
as being very similar to a multiple regression.
348
00:17:16.970 --> 00:17:18.830
So the big takeaway on this slide
349
00:17:18.830 --> 00:17:21.200
is that we see that genetic liability
350
00:17:21.200 --> 00:17:22.640
for cannabis use disorder
351
00:17:22.640 --> 00:17:25.360
is significantly positively associated
352
00:17:25.360 --> 00:17:28.900
with genetic risk for schizophrenia, even after we account
353
00:17:28.900 --> 00:17:31.615
for the effects of cannabis ever-use nicotine dependence
354
00:17:31.615 --> 00:17:33.800
and ever-smoking, as well
355
00:17:33.800 --> 00:17:36.980
as the correlations amongst these different phenotypes.
356
00:17:36.980 --> 00:17:37.890
I also wanna point out
357
00:17:37.890 --> 00:17:41.570
this interesting counterintuitive negative association
358
00:17:41.570 --> 00:17:44.550
that we see between ever smoking and schizophrenia.
359
00:17:44.550 --> 00:17:47.110
I don't really have time to talk about this today,
360
00:17:47.110 --> 00:17:50.210
but just keep in mind that that is effect that we see
361
00:17:50.210 --> 00:17:53.740
after partialing out the effects of these other substances.
362
00:17:53.740 --> 00:17:55.680
And if you're interested in talking about that more,
363
00:17:55.680 --> 00:17:57.710
please feel free to email me.
364
00:17:57.710 --> 00:17:58.973
On the next slide,
365
00:18:00.020 --> 00:18:02.150
we looked at this a slightly different way as well.
366
00:18:02.150 --> 00:18:04.720
So this is a slightly more conservative test.
367
00:18:04.720 --> 00:18:07.480
So here I've estimated this latent factor
368
00:18:07.480 --> 00:18:08.936
what I'm calling it
369
00:18:08.936 --> 00:18:09.860
cannabis and tobacco factor,
370
00:18:09.860 --> 00:18:12.580
and this really just represents the common genetic variance
371
00:18:12.580 --> 00:18:15.540
that's shared by a cannabis ever-use nicotine dependence
372
00:18:15.540 --> 00:18:16.980
and ever smoking.
373
00:18:16.980 --> 00:18:19.290
And what we see is that when we regressed schizophrenia
374
00:18:19.290 --> 00:18:22.930
onto cannabis use disorder and that latent factor,
375
00:18:22.930 --> 00:18:25.650
despite a really high significant correlation
376
00:18:25.650 --> 00:18:27.850
between cannabis use disorder and that latent factor,
377
00:18:27.850 --> 00:18:29.900
which we would expect, we see
378
00:18:29.900 --> 00:18:31.915
that cannabis use disorder
379
00:18:31.915 --> 00:18:33.640
is still significantly uniquely associated
380
00:18:33.640 --> 00:18:35.753
with schizophrenia risk.
381
00:18:36.720 --> 00:18:37.903
Next slide please.
382
00:18:39.550 --> 00:18:40.832
So here I'm gonna present
383
00:18:40.832 --> 00:18:43.380
to you the results of our variant-level analysis.
384
00:18:43.380 --> 00:18:44.950
So this is a Manhattan plot
385
00:18:44.950 --> 00:18:46.600
from our cross-disorder GWAS
386
00:18:46.600 --> 00:18:49.060
of cannabis use disorder and schizophrenia.
387
00:18:49.060 --> 00:18:52.470
Briefly, each of those dots represents a genetic variant
388
00:18:52.470 --> 00:18:53.950
in our analysis.
389
00:18:53.950 --> 00:18:56.420
And you can see that these are laid out along the x-axis
390
00:18:56.420 --> 00:18:59.170
by their physical position across the genome.
391
00:18:59.170 --> 00:19:01.870
On the y-axis, we have the negative log 10 P value,
392
00:19:01.870 --> 00:19:03.700
and that dash line represents
393
00:19:03.700 --> 00:19:06.220
our genome-wide significance threshold.
394
00:19:06.220 --> 00:19:07.661
So you can see
395
00:19:07.661 --> 00:19:09.490
that the strongest association here is actually
396
00:19:09.490 --> 00:19:10.380
on chromosome six
397
00:19:10.380 --> 00:19:13.680
at the major histocompatibility complex region,
398
00:19:13.680 --> 00:19:15.980
but I'm not really gonna talk about that today.
399
00:19:15.980 --> 00:19:18.840
It's most likely being driven by schizophrenia.
400
00:19:18.840 --> 00:19:21.180
In addition, that region has very complex linkage
401
00:19:21.180 --> 00:19:22.300
disequilibrium patterns,
402
00:19:22.300 --> 00:19:25.220
so it's a little difficult to parse what's going on there.
403
00:19:25.220 --> 00:19:27.780
So I'll focus on this chromosome eight locus today
404
00:19:27.780 --> 00:19:32.150
which is the second most significant association we see.
405
00:19:32.150 --> 00:19:33.820
And on the next slide,
406
00:19:33.820 --> 00:19:35.610
we can look at this a little bit more closely.
407
00:19:35.610 --> 00:19:39.660
So this is a zoomed-in regional plot of that locus.
408
00:19:39.660 --> 00:19:42.700
And we see that that top lead snip is an EQTL
409
00:19:42.700 --> 00:19:45.040
for several genes in this region.
410
00:19:45.040 --> 00:19:46.590
In particular, I wanna point out two
411
00:19:46.590 --> 00:19:49.053
of interest to CHRNA2s
412
00:19:49.053 --> 00:19:51.220
and and nicotinic acetylcholine receptor gene.
413
00:19:51.220 --> 00:19:53.470
It's been previously identified in GWAS
414
00:19:53.470 --> 00:19:56.166
of cannabis use disorder and schizophrenia.
415
00:19:56.166 --> 00:19:58.280
EPHX2 is also interesting.
416
00:19:58.280 --> 00:20:00.620
This gene is potentially involved
417
00:20:00.620 --> 00:20:02.850
in the metabolism of endocannabinoids.
418
00:20:02.850 --> 00:20:04.900
It's also involved in inflammation,
419
00:20:04.900 --> 00:20:06.040
and there's some evidence
420
00:20:06.040 --> 00:20:07.843
that it may play a role
421
00:20:07.843 --> 00:20:10.100
in neuro developmental disorders as well.
422
00:20:10.100 --> 00:20:13.670
Finally, on our next slide, in summary,
423
00:20:13.670 --> 00:20:16.330
we see the genetic liability for cannabis use disorder
424
00:20:16.330 --> 00:20:19.060
is uniquely significantly possibly associated
425
00:20:19.060 --> 00:20:21.850
with schizophrenia, even after we account for the effects
426
00:20:21.850 --> 00:20:24.243
of cannabis ever use and tobacco phenotypes.
427
00:20:25.140 --> 00:20:27.870
This chromosome eight locus may be one key point
428
00:20:27.870 --> 00:20:29.280
of shared genetic vulnerability,
429
00:20:29.280 --> 00:20:31.540
although I have to point out here, that doesn't mean
430
00:20:31.540 --> 00:20:33.650
that we expect that locus to explain a large amount
431
00:20:33.650 --> 00:20:36.140
of variance, these are highly apologetic traits.
432
00:20:36.140 --> 00:20:38.220
It may just be one really interesting place
433
00:20:38.220 --> 00:20:40.000
to follow up in the future.
434
00:20:40.000 --> 00:20:41.660
And finally, in terms of next steps,
435
00:20:41.660 --> 00:20:44.070
I think it will be interesting to see whether PRS
436
00:20:44.070 --> 00:20:46.430
from our cross-disorder GWAS predict the development
437
00:20:46.430 --> 00:20:48.121
of relevant behaviors in Longitudinal
438
00:20:48.121 --> 00:20:50.853
and high-risk datasets.
439
00:20:52.110 --> 00:20:55.540
Finally, I'd like to thank a few people, my coauthors
440
00:20:55.540 --> 00:20:59.273
and other analysts in the PGC SUD Group and Funding Support.
441
00:21:03.080 --> 00:21:06.770
Okay, do we have any questions for Emma right now?
442
00:21:06.770 --> 00:21:07.900
You can save them also.
443
00:21:07.900 --> 00:21:09.900
We should have a little time at the end.
444
00:21:11.271 --> 00:21:14.460
If not, then I think we should move
445
00:21:14.460 --> 00:21:17.520
on to our next speaker,
446
00:21:17.520 --> 00:21:22.520
which is Dr. Hollis Karoly from Colorado State University.
447
00:21:23.060 --> 00:21:24.283
Take it away, please.
448
00:21:25.910 --> 00:21:27.160
All right, well, thank you so much
449
00:21:27.160 --> 00:21:29.760
for giving me the opportunity to be here today.
450
00:21:29.760 --> 00:21:30.770
I'm Hollis Karoly,
451
00:21:30.770 --> 00:21:33.270
I'm the new Assistant Professor at Colorado State,
452
00:21:33.270 --> 00:21:35.030
and I'm gonna be telling you today
453
00:21:35.030 --> 00:21:36.400
about a project that I began
454
00:21:36.400 --> 00:21:39.100
during my postdoc at the University of Colorado.
455
00:21:39.100 --> 00:21:42.470
And in this project, we looked at relationships
456
00:21:42.470 --> 00:21:45.280
between neural structure using neuroimaging measures,
457
00:21:45.280 --> 00:21:47.310
peripheral cytokines and a protein
458
00:21:47.310 --> 00:21:49.270
called neurofilament light
459
00:21:49.270 --> 00:21:51.800
in a sample of heavy drinkers.
460
00:21:51.800 --> 00:21:53.270
Next slide.
461
00:21:53.270 --> 00:21:56.200
So excess alcohol consumption has been found
462
00:21:56.200 --> 00:21:58.760
to cause inflammation within the body and brain.
463
00:21:58.760 --> 00:22:01.360
And unfortunately, it's difficult to measure inflammation
464
00:22:01.360 --> 00:22:02.810
in the brain in living humans.
465
00:22:02.810 --> 00:22:04.800
So human researchers like myself
466
00:22:04.800 --> 00:22:07.330
often use peripheral inflammatory markers
467
00:22:07.330 --> 00:22:09.590
as sort of a proxy for general inflammation.
468
00:22:09.590 --> 00:22:11.830
And that's what we've done in this study.
469
00:22:11.830 --> 00:22:14.162
It's also known that alcohol and inflammation,
470
00:22:14.162 --> 00:22:18.330
or data suggests that alcohol and inflammation likely impact
471
00:22:18.330 --> 00:22:19.797
both gray and white matter in the brain.
472
00:22:19.797 --> 00:22:23.440
And some studies have indicated that frontal brain regions,
473
00:22:23.440 --> 00:22:25.360
those that subserve executive function,
474
00:22:25.360 --> 00:22:29.130
may be particularly vulnerable to this type of damage.
475
00:22:29.130 --> 00:22:32.220
So more recent research has indicated that damaged cells
476
00:22:32.220 --> 00:22:35.360
in the brain release a protein
477
00:22:35.360 --> 00:22:37.527
called neurofilament light.
478
00:22:37.527 --> 00:22:40.140
Neurofilament light is a neurofilament protein.
479
00:22:40.140 --> 00:22:42.880
It's one of five neurofilament proteins that act
480
00:22:42.880 --> 00:22:45.610
to provide structural support for neurons.
481
00:22:45.610 --> 00:22:48.510
And following accidental damage or neuro-degeneration,
482
00:22:48.510 --> 00:22:50.320
these neurofilament proteins are released
483
00:22:50.320 --> 00:22:52.320
into the interstitial space,
484
00:22:52.320 --> 00:22:55.150
into the CSF and ultimately into blood.
485
00:22:55.150 --> 00:22:58.380
Now, NfL is the most abundant of the neurofilament proteins
486
00:22:58.380 --> 00:23:00.260
and it's relatively easily measured
487
00:23:00.260 --> 00:23:03.120
using just like a standard Eliza assay,
488
00:23:03.120 --> 00:23:05.310
and so for these reasons, NfL has emerged
489
00:23:05.310 --> 00:23:07.900
in recent years as a promising potential marker
490
00:23:07.900 --> 00:23:10.620
for a number of neurodegenerative disorders.
491
00:23:10.620 --> 00:23:12.810
To date however, it has not been explored
492
00:23:12.810 --> 00:23:14.670
in the context of alcohol use disorders,
493
00:23:14.670 --> 00:23:17.560
and that was what we were particularly interested in.
494
00:23:17.560 --> 00:23:19.140
So in this study, I'm gonna be talking
495
00:23:19.140 --> 00:23:21.320
about just exploring some relationships
496
00:23:21.320 --> 00:23:24.570
between alcohol consumption, peripheral inflammation,
497
00:23:24.570 --> 00:23:28.030
gray and white matter in the brain and neurofilament light.
498
00:23:28.030 --> 00:23:28.973
Next slide please.
499
00:23:30.900 --> 00:23:34.460
So today I'll be telling you about data from 74 participants
500
00:23:34.460 --> 00:23:37.969
who were a subset of a larger clinical intervention study
501
00:23:37.969 --> 00:23:40.510
for heavy drinkers who wanted to cut down
502
00:23:40.510 --> 00:23:42.240
on their alcohol consumption.
503
00:23:42.240 --> 00:23:43.073
And all the data
504
00:23:43.073 --> 00:23:44.700
that I'm gonna be presenting today comes
505
00:23:44.700 --> 00:23:47.780
from baseline session that these participants underwent
506
00:23:47.780 --> 00:23:50.360
prior to actually receiving the intervention.
507
00:23:50.360 --> 00:23:54.080
So at baseline, participants underwent an MRI scan
508
00:23:54.080 --> 00:23:56.710
where we ran structural sequences to collect
509
00:23:56.710 --> 00:23:59.060
gray matter data and diffusion sequences
510
00:23:59.060 --> 00:24:00.750
to get white matter data.
511
00:24:00.750 --> 00:24:03.580
We also obtained a blood sample from our participants.
512
00:24:03.580 --> 00:24:05.700
And so from this blood sample, we were able
513
00:24:05.700 --> 00:24:09.230
to measure the pro-inflammatory cytokine IL-6.
514
00:24:09.230 --> 00:24:11.080
Again, that's just our peripheral kind
515
00:24:11.080 --> 00:24:13.480
of general inflammatory marker
516
00:24:13.480 --> 00:24:16.880
since we can't easily measure inflammation in the brain.
517
00:24:16.880 --> 00:24:18.660
And then also from the blood samples,
518
00:24:18.660 --> 00:24:22.110
we are measuring NfL, the protein that I just mentioned.
519
00:24:22.110 --> 00:24:23.060
Next slide, please.
520
00:24:24.680 --> 00:24:27.460
So we hypothesized that greater alcohol consumption
521
00:24:27.460 --> 00:24:29.900
in this sample would be associated with higher levels
522
00:24:29.900 --> 00:24:32.960
of circulating IL-6, lower gray matter thickness
523
00:24:32.960 --> 00:24:35.090
and higher white matter diffusivity.
524
00:24:35.090 --> 00:24:37.510
And just as a reminder, low gray matter thickness
525
00:24:37.510 --> 00:24:39.920
and higher white matter diffusivity are sort of indicative
526
00:24:39.920 --> 00:24:41.083
of neural damage.
527
00:24:42.060 --> 00:24:44.910
We also hypothesized that circulating IL-6
528
00:24:44.910 --> 00:24:47.360
would also be associated with lower gray matter thickness
529
00:24:47.360 --> 00:24:49.560
and higher white matter diffusivity.
530
00:24:49.560 --> 00:24:51.930
And that these two markers of neural damage
531
00:24:51.930 --> 00:24:53.250
would both be associated
532
00:24:53.250 --> 00:24:56.100
with increased levels of circulating neurofilament light.
533
00:24:57.220 --> 00:24:58.053
Next slide.
534
00:24:59.680 --> 00:25:02.010
So we extracted gray matter thickness values
535
00:25:02.010 --> 00:25:04.880
from the bilateral rostral and caudal middle frontal regions
536
00:25:04.880 --> 00:25:06.410
of the brain using free surfer.
537
00:25:06.410 --> 00:25:07.690
Again, just to kind of focus
538
00:25:07.690 --> 00:25:09.993
in on those interesting frontal regions.
539
00:25:10.850 --> 00:25:13.420
We also calculated axial radial
540
00:25:13.420 --> 00:25:16.280
and mean diffusivity using track-based spatial statistics
541
00:25:16.280 --> 00:25:18.250
and extracted those values,
542
00:25:18.250 --> 00:25:21.530
and then created a composite score composed of a number
543
00:25:21.530 --> 00:25:23.680
of different white matter tracks that have been associated
544
00:25:23.680 --> 00:25:27.250
with alcohol use in prior studies.
545
00:25:27.250 --> 00:25:28.660
We ran some regression models
546
00:25:28.660 --> 00:25:32.260
in which each outcome variable of interest was regressed
547
00:25:32.260 --> 00:25:34.630
on the predictor of interest as well as gender,
548
00:25:34.630 --> 00:25:37.320
and for all the models that involved gray matter,
549
00:25:37.320 --> 00:25:40.220
we also included total estimated intercranial volume
550
00:25:40.220 --> 00:25:41.263
as a covariate.
551
00:25:42.200 --> 00:25:44.240
Finally as sort of an exploratory addition
552
00:25:44.240 --> 00:25:45.910
to this project, we followed up
553
00:25:45.910 --> 00:25:48.400
on significant white and gray matter results
554
00:25:48.400 --> 00:25:49.760
using just a whole-brain approach
555
00:25:49.760 --> 00:25:53.700
to kind of see how these relationships played out
556
00:25:53.700 --> 00:25:54.690
across the entire brain
557
00:25:54.690 --> 00:25:56.360
and not just in those particular tracks
558
00:25:56.360 --> 00:25:58.160
or regions of interest.
559
00:25:58.160 --> 00:25:58.993
Next slide.
560
00:26:00.540 --> 00:26:03.300
So overall, these regression models supported
561
00:26:03.300 --> 00:26:06.950
each all of our hypothesis except for the relationship
562
00:26:06.950 --> 00:26:09.780
between circulating IL-6 and white matter.
563
00:26:09.780 --> 00:26:10.840
So we actually found
564
00:26:10.840 --> 00:26:13.850
that there was no association between IL-6
565
00:26:13.850 --> 00:26:16.510
and white matter diffusivity in this sample.
566
00:26:16.510 --> 00:26:17.343
Next slide.
567
00:26:19.550 --> 00:26:21.870
And you can check out my poster for more detail
568
00:26:21.870 --> 00:26:25.940
on these results, but this is just a table that kind
569
00:26:25.940 --> 00:26:27.820
of shows the overall regression models.
570
00:26:27.820 --> 00:26:30.408
You can see that gender was included as a covariate
571
00:26:30.408 --> 00:26:34.500
as well as in the gray matter models,
572
00:26:34.500 --> 00:26:37.010
we also included intercranial volume.
573
00:26:37.010 --> 00:26:37.843
Next slide.
574
00:26:40.280 --> 00:26:43.070
So results of our whole-brain TBSS analysis showed
575
00:26:43.070 --> 00:26:47.030
that both NfL and percent drinking days was associated
576
00:26:47.030 --> 00:26:50.010
with greater diffusivity across the number of tracks.
577
00:26:50.010 --> 00:26:51.830
One of the things that I thought was kind of interesting
578
00:26:51.830 --> 00:26:53.420
about these results is that the effects
579
00:26:53.420 --> 00:26:56.310
of NfL was pretty strong, passed a threshold
580
00:26:57.924 --> 00:26:59.020
of P less than 0.05, but we didn't find this effect
581
00:26:59.020 --> 00:27:02.680
in the percent heavy drinking days analysis.
582
00:27:02.680 --> 00:27:04.040
So that was a bit surprising.
583
00:27:04.040 --> 00:27:05.210
And I will comment on that
584
00:27:05.210 --> 00:27:06.670
in the discussion section.
585
00:27:06.670 --> 00:27:07.503
Next slide.
586
00:27:09.530 --> 00:27:12.540
So results of from our gray matter whole-brain analysis
587
00:27:12.540 --> 00:27:14.930
also showed that heavy drinking days, NfL
588
00:27:14.930 --> 00:27:16.500
and IL-6 were all associated
589
00:27:16.500 --> 00:27:18.800
with decreased cortical thickness across a number
590
00:27:18.800 --> 00:27:21.890
of regions, including the frontal regions of interest.
591
00:27:21.890 --> 00:27:24.322
And you can take a closer look at this on my poster.
592
00:27:24.322 --> 00:27:25.563
Next slide.
593
00:27:28.360 --> 00:27:30.100
So overall, these results suggest
594
00:27:30.100 --> 00:27:32.550
that alcohol is associated with inflammation
595
00:27:32.550 --> 00:27:35.400
which is linked with structural brain damage and NfL.
596
00:27:35.400 --> 00:27:37.670
This is the first study to show an association
597
00:27:37.670 --> 00:27:40.590
between brain structure and NfL in heavy drinkers.
598
00:27:40.590 --> 00:27:43.020
And although a lot of research should still be done
599
00:27:43.020 --> 00:27:45.590
on this as a biomarker, this work does suggest
600
00:27:45.590 --> 00:27:48.760
that NfL may be a useful biomarker when it comes
601
00:27:48.760 --> 00:27:52.100
to neural damage in samples of heavy drinkers,
602
00:27:52.100 --> 00:27:54.490
and it may have potentially some clinical implications
603
00:27:54.490 --> 00:27:56.100
down the line.
604
00:27:56.100 --> 00:27:57.710
We failed to observe an association
605
00:27:57.710 --> 00:27:59.510
between white matter and IL-6,
606
00:27:59.510 --> 00:28:02.440
and it could be because there is just no association
607
00:28:02.440 --> 00:28:04.670
between peripheral cytokines and white matter,
608
00:28:04.670 --> 00:28:07.410
or perhaps IL-6 alone is not an adequate marker.
609
00:28:07.410 --> 00:28:09.920
So maybe a composite of cytokines would be
610
00:28:09.920 --> 00:28:12.090
a more useful inflammatory marker
611
00:28:12.090 --> 00:28:13.623
for future studies like this.
612
00:28:14.690 --> 00:28:16.930
We also saw a weaker-than-expected association
613
00:28:16.930 --> 00:28:18.860
between drinking and white matter in this study.
614
00:28:18.860 --> 00:28:19.950
And I think this was likely
615
00:28:19.950 --> 00:28:23.680
because the heavy-drinking individuals that we included
616
00:28:23.680 --> 00:28:25.910
in this study were not as severe drinkers
617
00:28:25.910 --> 00:28:29.290
as those that our group
618
00:28:30.596 --> 00:28:32.070
and others have looked at in conjunction with white matter
619
00:28:32.070 --> 00:28:33.070
in previous research.
620
00:28:33.070 --> 00:28:36.060
And so I think that this just necessitates some replication
621
00:28:36.060 --> 00:28:38.760
in more severe drinking samples.
622
00:28:38.760 --> 00:28:39.593
Next slide.
623
00:28:41.540 --> 00:28:43.970
So I'd like to thank our funders at NIAAA
624
00:28:43.970 --> 00:28:45.390
as well as my mentors
625
00:28:46.419 --> 00:28:47.870
and collaborators from the University of Colorado Boulder
626
00:28:47.870 --> 00:28:50.593
as well as my current institution, Colorado State.
627
00:28:52.420 --> 00:28:53.470
Okay, thank you.
628
00:28:53.470 --> 00:28:56.310
We have a couple questions in for you.
629
00:28:56.310 --> 00:28:57.450
Okay.
630
00:28:57.450 --> 00:28:59.040
I'll read them.
631
00:28:59.040 --> 00:29:03.170
NfLs are also biomarkers for peripheral neuropathy.
632
00:29:03.170 --> 00:29:05.485
Did you assess these patients
633
00:29:05.485 --> 00:29:10.485
if they had alcohol-induced problems neuropathy?
634
00:29:12.360 --> 00:29:14.510
So that's a fantastic point.
635
00:29:14.510 --> 00:29:16.690
We did not in this particular study,
636
00:29:16.690 --> 00:29:18.980
but I think that would be an important future direction.
637
00:29:18.980 --> 00:29:22.780
And yeah, thank you for the comment.
638
00:29:22.780 --> 00:29:24.980
That's a great idea, a great thought.
639
00:29:24.980 --> 00:29:26.880
Okay, and another question that came in
640
00:29:26.880 --> 00:29:31.500
in our chat was how do you distinguish the NfL release
641
00:29:31.500 --> 00:29:36.280
due to neuro-degeneration or other brain damages
642
00:29:36.280 --> 00:29:38.137
versus those from AUD?
643
00:29:39.400 --> 00:29:41.320
So I think that that's something
644
00:29:41.320 --> 00:29:43.770
that we can't necessarily distinguish at this point.
645
00:29:43.770 --> 00:29:47.750
All of that the NfL tells us is that individuals
646
00:29:47.750 --> 00:29:49.310
may have some neural damage.
647
00:29:49.310 --> 00:29:52.060
And in the context of folks with AUD,
648
00:29:52.060 --> 00:29:52.893
we don't know
649
00:29:52.893 --> 00:29:55.380
whether their neural damage is 100% due to their drinking.
650
00:29:55.380 --> 00:29:57.790
It could be due to other stressors in your life
651
00:29:57.790 --> 00:30:02.790
or other medical conditions, other substance use.
652
00:30:03.520 --> 00:30:05.520
So I think that there's still a lot
653
00:30:05.520 --> 00:30:06.810
of research that needs to be done
654
00:30:06.810 --> 00:30:08.990
to figure out how best we can use NfL
655
00:30:08.990 --> 00:30:12.810
as a marker in the context of AUD and other disorders.
656
00:30:12.810 --> 00:30:15.597
But that's a great point and one that definitely kind
657
00:30:15.597 --> 00:30:18.683
of stumped us when we were working on this project as well.
658
00:30:19.907 --> 00:30:21.560
Okay, we've got a couple more questions in
659
00:30:21.560 --> 00:30:23.300
and we still have some time.
660
00:30:23.300 --> 00:30:25.560
Can you visualize NfLs
661
00:30:25.560 --> 00:30:30.410
by regular immuno processes in postmortem tissue?
662
00:30:30.410 --> 00:30:34.583
And if so, what regional differences would you expect?
663
00:30:35.840 --> 00:30:37.360
That's a great question.
664
00:30:37.360 --> 00:30:38.560
So to my knowledge,
665
00:30:38.560 --> 00:30:40.391
I haven't read any studies that have done that,
666
00:30:40.391 --> 00:30:42.969
but I'm not 100% sure.
667
00:30:42.969 --> 00:30:45.700
I would have to pop back
668
00:30:45.700 --> 00:30:48.210
and look at the literature to make sure
669
00:30:48.210 --> 00:30:50.750
that I'm giving you the right answer to that question.
670
00:30:50.750 --> 00:30:51.583
Yeah, sure.
671
00:30:51.583 --> 00:30:53.550
And then we got another one coming in,
672
00:30:53.550 --> 00:30:58.403
is plasma NfL biomarker for neurological disease?
673
00:30:59.350 --> 00:31:01.970
So at this point, that's how it's being used.
674
00:31:01.970 --> 00:31:05.170
So it's been used as a marker, like for example,
675
00:31:05.170 --> 00:31:09.762
in Alzheimer's research or in the context of MS and ALS,
676
00:31:09.762 --> 00:31:12.250
they are collecting plasma levels
677
00:31:12.250 --> 00:31:14.150
of NfL and using it as a marker.
678
00:31:14.150 --> 00:31:15.870
So we were sort of following
679
00:31:15.870 --> 00:31:18.530
what's being done in the context
680
00:31:18.530 --> 00:31:19.900
of these other disorders.
681
00:31:19.900 --> 00:31:21.010
So yeah.
682
00:31:21.010 --> 00:31:22.324
Okay, good.
683
00:31:22.324 --> 00:31:24.130
And then we still got a few more minutes.
684
00:31:24.130 --> 00:31:29.130
So next chat says, I may have missed this, in your model,
685
00:31:29.680 --> 00:31:32.450
what are you considering heavy drinking days?
686
00:31:32.450 --> 00:31:34.923
Have you looked at intermittent binge drinking?
687
00:31:35.910 --> 00:31:39.230
So in this model we looked at, and I didn't go over this.
688
00:31:39.230 --> 00:31:42.560
You should look at my poster too for more detail.
689
00:31:42.560 --> 00:31:45.020
But we were looking at percentage heavy drinking days.
690
00:31:45.020 --> 00:31:46.638
So days where they drank five drinks
691
00:31:46.638 --> 00:31:49.170
or more, the percentage
692
00:31:49.170 --> 00:31:51.226
of their total drinking days that were heavy drinking days.
693
00:31:51.226 --> 00:31:53.760
I also looked at total number of drinking days
694
00:31:53.760 --> 00:31:58.340
in the last week, I looked at drinks per drinking day.
695
00:31:58.340 --> 00:32:00.457
We tried looking at AUDIT scores as well,
696
00:32:00.457 --> 00:32:03.750
but it's so, and we also tried
697
00:32:03.750 --> 00:32:04.877
like some lane variables across all of these.
698
00:32:04.877 --> 00:32:08.820
So yeah, I think that there's potentially more exploration
699
00:32:08.820 --> 00:32:11.210
around which alcohol measures are best.
700
00:32:11.210 --> 00:32:13.810
But in general, percent heavy drinking days tends
701
00:32:13.810 --> 00:32:16.960
to work pretty well for us in these types of projects.
702
00:32:16.960 --> 00:32:19.200
So that's kind of what we went with.
703
00:32:19.200 --> 00:32:21.110
Okay, I think we can get one more question
704
00:32:21.110 --> 00:32:24.110
and then we'll maybe try to get to some of these at the end.
705
00:32:25.460 --> 00:32:27.820
Your talk certainly generated a lot of interest.
706
00:32:27.820 --> 00:32:32.820
So let's do this last question and we'll move on.
707
00:32:32.900 --> 00:32:35.700
Does NfL release into periphery?
708
00:32:35.700 --> 00:32:39.493
If so, can you measure it in blood as a bio mark?
709
00:32:40.806 --> 00:32:43.350
So my understanding is that NfL is released
710
00:32:43.350 --> 00:32:45.980
by following neural damage in the brain.
711
00:32:45.980 --> 00:32:50.290
So it goes into the CSF and then into the blood.
712
00:32:50.290 --> 00:32:52.860
And then we're using that as a biomarker
713
00:32:52.860 --> 00:32:54.460
of what's going on in the brain.
714
00:32:55.490 --> 00:32:56.659
Okay.
715
00:32:56.659 --> 00:32:57.663
Thank you very much.
716
00:32:57.663 --> 00:33:00.380
And it's time to move on to our next speaker
717
00:33:00.380 --> 00:33:03.700
who is Kathryn Biernacki,
718
00:33:03.700 --> 00:33:06.710
I may have mispronounced that, from Rutgers University,
719
00:33:06.710 --> 00:33:08.090
the Newark campus.
720
00:33:08.090 --> 00:33:09.900
Kathryn, take it away.
721
00:33:09.900 --> 00:33:10.890
Thank you.
722
00:33:10.890 --> 00:33:11.970
My name is Kathryn Biernacki.
723
00:33:11.970 --> 00:33:14.690
I'm a post-doc at Rutgers University Newark
724
00:33:14.690 --> 00:33:17.990
at the Center for Molecular and Behavioral Neuroscience.
725
00:33:17.990 --> 00:33:19.700
And today I'll be presenting a study
726
00:33:19.700 --> 00:33:21.730
looking at improving reward processing
727
00:33:21.730 --> 00:33:24.550
in substance users using neuro-modulation.
728
00:33:24.550 --> 00:33:25.980
Next slide, please.
729
00:33:25.980 --> 00:33:28.010
Theoretical and empirical work suggests
730
00:33:28.010 --> 00:33:30.230
that the anterior cingulate cortex uses
731
00:33:30.230 --> 00:33:32.940
dopaminergic reward prediction error signals to learn
732
00:33:32.940 --> 00:33:36.060
the value of rewards to motivate goal-directed behavior.
733
00:33:36.060 --> 00:33:38.280
And research shows us that singular function
734
00:33:38.280 --> 00:33:40.600
is impaired in substance use disorder.
735
00:33:40.600 --> 00:33:41.750
And this impairment contributes
736
00:33:41.750 --> 00:33:43.570
to the abnormal cognitive control
737
00:33:43.570 --> 00:33:45.060
and decision-making processes
738
00:33:45.060 --> 00:33:46.843
we see in substance use disorder.
739
00:33:48.130 --> 00:33:49.150
Next slide, please.
740
00:33:49.150 --> 00:33:50.350
The way we can measure function
741
00:33:50.350 --> 00:33:52.120
of the anterior cingulate cortex
742
00:33:52.120 --> 00:33:54.580
is via an electric physiological signal
743
00:33:54.580 --> 00:33:56.420
called the reward positivity.
744
00:33:56.420 --> 00:33:58.170
This is a positive-going deflection
745
00:33:58.170 --> 00:34:01.790
in the AHA elicited by reward-related neural-processes.
746
00:34:01.790 --> 00:34:04.260
And we calculate it by looking at the difference
747
00:34:04.260 --> 00:34:05.920
between responses to rewards and neuro-rewards.
748
00:34:05.920 --> 00:34:09.230
So the reward positivity reflects the receipt
749
00:34:09.230 --> 00:34:11.400
of reward prediction error signals carried
750
00:34:11.400 --> 00:34:13.600
by the midbrain dopamine system
751
00:34:13.600 --> 00:34:16.620
to the anterior mid-singular cortex, which then uses this
752
00:34:16.620 --> 00:34:19.740
for adoptive modification of goal-directed behavior.
753
00:34:19.740 --> 00:34:21.300
So we can use this signal to analyze
754
00:34:21.300 --> 00:34:24.100
impaired reward processing in substance use disorder.
755
00:34:24.100 --> 00:34:26.470
And indeed our team has found
756
00:34:26.470 --> 00:34:28.700
that the reward positivity is dampened
757
00:34:28.700 --> 00:34:30.450
in response to monitor your rewards
758
00:34:30.450 --> 00:34:34.780
in nicotine and substance uses relative to healthy controls.
759
00:34:34.780 --> 00:34:35.840
Next slide.
760
00:34:35.840 --> 00:34:36.673
We can also look
761
00:34:36.673 --> 00:34:39.100
at reward processing in substance use disorder
762
00:34:39.100 --> 00:34:42.950
by a striatal processing of reward prediction error signals.
763
00:34:42.950 --> 00:34:45.440
A biologically-biased model of reinforcement learning
764
00:34:45.440 --> 00:34:48.130
calls that positive reward prediction error signals,
765
00:34:48.130 --> 00:34:49.540
facilitate approach learning
766
00:34:51.452 --> 00:34:52.509
while negative reward prediction error signals
767
00:34:52.509 --> 00:34:53.342
facilitate an avoidance learning
768
00:34:53.342 --> 00:34:56.580
by a striatal D1 and D2 receptors respectively.
769
00:34:56.580 --> 00:34:58.010
An empirical way of measuring this
770
00:34:58.010 --> 00:35:00.223
is by the probabilistic selection task.
771
00:35:01.090 --> 00:35:03.410
This task is sensitive to dopamine dysfunction.
772
00:35:03.410 --> 00:35:04.980
And given that most drugs of abuse
773
00:35:04.980 --> 00:35:08.290
modulate the magnitude of reward prediction error signals,
774
00:35:08.290 --> 00:35:10.980
we hypothesized that drug use might also affect
775
00:35:10.980 --> 00:35:13.540
PST performance giving us a behavioral measure
776
00:35:13.540 --> 00:35:15.293
of impaired reward function.
777
00:35:16.450 --> 00:35:17.283
Next slide.
778
00:35:17.283 --> 00:35:18.410
So how do we counteract
779
00:35:19.542 --> 00:35:20.880
these drug-induced neural-cognitive deficits?
780
00:35:20.880 --> 00:35:22.670
Well, an FDA-approved treatment
781
00:35:22.670 --> 00:35:24.910
for depression might give us the solution.
782
00:35:24.910 --> 00:35:27.760
10 Hertz repetitive transcranial magnetic stimulation,
783
00:35:27.760 --> 00:35:31.210
or TMS to the left also lateral prefrontal cortex,
784
00:35:31.210 --> 00:35:33.250
has been shown to enhance dopamine release,
785
00:35:33.250 --> 00:35:35.930
neuronal activity and cerebral blood flow
786
00:35:35.930 --> 00:35:38.820
in the singular cortex of healthy individuals.
787
00:35:38.820 --> 00:35:40.870
And although the anterior singular cortex
788
00:35:40.870 --> 00:35:43.740
is too deeply located for TMS to directly modulate
789
00:35:43.740 --> 00:35:45.142
its activity, we can modulate its activity
790
00:35:45.142 --> 00:35:46.405
by the DLPFC
791
00:35:46.405 --> 00:35:48.679
because of dense reciprocal anatomical connections
792
00:35:48.679 --> 00:35:51.480
between the two.
793
00:35:51.480 --> 00:35:52.420
And it's been suggested
794
00:35:52.420 --> 00:35:54.450
that this non-invasive brain stimulation method is capable
795
00:35:54.450 --> 00:35:58.530
of modulating reward prediction error-related activity
796
00:35:58.530 --> 00:36:01.470
in the anterior singular cortex and the striatum,
797
00:36:01.470 --> 00:36:04.414
and our research has showing the TMS is capable
798
00:36:04.414 --> 00:36:06.830
of increasing the reward positivity
799
00:36:06.830 --> 00:36:09.273
in response to monetary reward in smokers.
800
00:36:10.590 --> 00:36:11.866
Next slide.
801
00:36:11.866 --> 00:36:12.699
So pulling all of this together,
802
00:36:12.699 --> 00:36:14.610
in this proof-of-concept study, we aim to use
803
00:36:14.610 --> 00:36:18.050
10 Hertz excitatory robot-assisted TMS to enhance
804
00:36:18.050 --> 00:36:20.350
reward processing in substance use disorder.
805
00:36:20.350 --> 00:36:22.690
And we aim to do this by modulating the amplitude
806
00:36:22.690 --> 00:36:25.340
of the reward positivity as well as approach learning
807
00:36:27.195 --> 00:36:28.028
as measured by the PST.
808
00:36:28.028 --> 00:36:30.580
We collected data on 22 substance users who used a range
809
00:36:30.580 --> 00:36:32.230
of substances and they were randomized
810
00:36:32.230 --> 00:36:34.763
into either active TMS or SHAM, sorry, placebo.
811
00:36:36.910 --> 00:36:39.865
Briefly, we applied 10 Hertz excitatory TMS to the DLPFC
812
00:36:39.865 --> 00:36:41.960
and we used a robot that allowed us
813
00:36:41.960 --> 00:36:45.650
to precisely position the TMS coil over our S3 target
814
00:36:45.650 --> 00:36:48.290
where the DLPFC is believed to be.
815
00:36:48.290 --> 00:36:49.430
Based on individual maps
816
00:36:49.430 --> 00:36:51.310
that we created for each participant's head,
817
00:36:51.310 --> 00:36:53.340
which could then be used to track their head position
818
00:36:53.340 --> 00:36:56.470
in space and maintain target precision.
819
00:36:56.470 --> 00:36:58.430
Participants were saved TMS while they completed
820
00:36:58.430 --> 00:37:01.340
a TMS task by their saved rewards intervals,
821
00:37:01.340 --> 00:37:03.210
and they completed 100 trials of this task
822
00:37:03.210 --> 00:37:05.816
as a baseline without TMS, and then 300 trials with TMS.
823
00:37:05.816 --> 00:37:08.885
And the SHAM condition just involved flipping
824
00:37:08.885 --> 00:37:11.680
the TMS coil, so participants were saved
825
00:37:11.680 --> 00:37:13.150
the same auditory stimulation
826
00:37:13.150 --> 00:37:15.601
but in our actual neurostimulation.
827
00:37:15.601 --> 00:37:17.570
And we measured their reward positivity as a difference
828
00:37:17.570 --> 00:37:20.660
between the reward and neuro-reward conditions in this task.
829
00:37:20.660 --> 00:37:22.100
And then they completed the PST
830
00:37:22.100 --> 00:37:24.163
immediately after the TMS task.
831
00:37:25.500 --> 00:37:26.407
Jumping into our results,
832
00:37:26.407 --> 00:37:29.100
our participants used a broad range of substances
833
00:37:29.100 --> 00:37:31.440
and that red line their indicates a problematic level
834
00:37:31.440 --> 00:37:34.050
of drug use in the assist drug screening questionnaire
835
00:37:34.050 --> 00:37:35.800
that we use to screen participants.
836
00:37:37.260 --> 00:37:38.623
But the reward positivity,
837
00:37:40.955 --> 00:37:41.790
we can say that overall, TMS was successful
838
00:37:43.335 --> 00:37:44.168
in modulating its amplitude and the active relative
839
00:37:44.168 --> 00:37:45.260
to the SHAM condition.
840
00:37:45.260 --> 00:37:47.820
When we look at AIPs across blocks, neither group
841
00:37:47.820 --> 00:37:50.090
produced a reward positivity for the first few blocks
842
00:37:50.090 --> 00:37:51.530
of the TMS task.
843
00:37:51.530 --> 00:37:54.090
But then we say the active has group developed
844
00:37:54.090 --> 00:37:56.890
a reward positivity in block three and four.
845
00:37:56.890 --> 00:37:59.160
Meanwhile, the SHAM group failed to reproduce
846
00:37:59.160 --> 00:38:01.150
a reward positivity at all.
847
00:38:01.150 --> 00:38:03.000
So we say that there's a cumulative effect
848
00:38:03.000 --> 00:38:04.970
of the active TMS and it really kicks in
849
00:38:04.970 --> 00:38:06.270
for blocks three and four.
850
00:38:07.630 --> 00:38:09.948
Looking at the reward learning on the PST,
851
00:38:09.948 --> 00:38:12.690
TMS also significantly modulated approach learning
852
00:38:12.690 --> 00:38:14.300
with participants in the active condition
853
00:38:14.300 --> 00:38:15.980
performing better than SHAM.
854
00:38:15.980 --> 00:38:17.870
Interestingly, its approach learning performance
855
00:38:17.870 --> 00:38:19.290
was correlated with the amplitude
856
00:38:19.290 --> 00:38:22.083
of the reward positivity in block four, the TMS task.
857
00:38:22.930 --> 00:38:25.490
Together, these results suggest that excitatory TMS
858
00:38:25.490 --> 00:38:27.570
can alleviate (indistinct) processing
859
00:38:27.570 --> 00:38:30.530
in the anterior singular cortex of substance users.
860
00:38:30.530 --> 00:38:31.884
And we say this through the amplification
861
00:38:31.884 --> 00:38:34.780
of the magnitude of the reward positivity.
862
00:38:34.780 --> 00:38:36.440
Behaviorally, we can say that TMS also has
863
00:38:36.440 --> 00:38:39.385
a significant impact on reward-related decision making.
864
00:38:39.385 --> 00:38:42.720
Pulling all of this together, TMS looks like it enhances
865
00:38:42.720 --> 00:38:45.140
dopaminergic reward prediction error signaling,
866
00:38:45.140 --> 00:38:47.530
following the theories we looked at earlier.
867
00:38:47.530 --> 00:38:49.470
Significantly, these results were promising
868
00:38:49.470 --> 00:38:52.130
for developing new avenues of addiction treatment,
869
00:38:52.130 --> 00:38:54.490
by using TMS to modulate reward processing
870
00:38:54.490 --> 00:38:55.790
and using the reward positivity
871
00:38:55.790 --> 00:38:58.423
and PST to monitor the effects of this treatment.
872
00:38:59.270 --> 00:39:01.565
Moving forward (indistinct) to conduct this study
873
00:39:01.565 --> 00:39:03.195
in an opioid using population,
874
00:39:03.195 --> 00:39:05.600
which is my specific group of interest and specialty,
875
00:39:05.600 --> 00:39:09.340
so we can look at substance-specific effects.
876
00:39:09.340 --> 00:39:10.670
We're also interested in determining
877
00:39:10.670 --> 00:39:12.410
whether we can boost the effect of TMS
878
00:39:12.410 --> 00:39:15.380
by using a more individualized target than F3.
879
00:39:15.380 --> 00:39:17.120
So we recently submitted a grant to say,
880
00:39:17.120 --> 00:39:19.820
we can find individualized DLPFC targets using measures
881
00:39:19.820 --> 00:39:20.820
like diffusion weighting
882
00:39:20.820 --> 00:39:22.590
and cortical thickness to really boost
883
00:39:22.590 --> 00:39:24.260
the reward positivity.
884
00:39:24.260 --> 00:39:26.230
The next logical step would also be to determine
885
00:39:26.230 --> 00:39:28.095
whether we can get long-term potentiation
886
00:39:28.095 --> 00:39:30.710
of these TMS effects over multiple sessions.
887
00:39:30.710 --> 00:39:32.980
And given that drug treatment is generally best supported
888
00:39:32.980 --> 00:39:35.140
by behavioral therapies, incorporating TMS
889
00:39:35.140 --> 00:39:37.040
with these therapies also makes sense.
890
00:39:37.040 --> 00:39:38.368
So that's why the (indistinct)
891
00:39:38.368 --> 00:39:41.090
to really expand the scope of this research.
892
00:39:41.090 --> 00:39:43.160
Given that it's still in the midst of an opioid crisis,
893
00:39:43.160 --> 00:39:45.580
I'm gonna put together a research program
894
00:39:45.580 --> 00:39:47.577
in opioid users that brings together
895
00:39:47.577 --> 00:39:49.970
longitudinal TMS research with behavioral therapy to say,
896
00:39:49.970 --> 00:39:51.180
if we can incorporate
897
00:39:52.255 --> 00:39:53.230
this noninvasive neuromodulatory method
898
00:39:53.230 --> 00:39:55.870
into standard treatment, which will then hopefully result
899
00:39:55.870 --> 00:39:57.920
in meaningful and sustained recovery
900
00:39:57.920 --> 00:39:59.170
from opioid use disorder.
901
00:40:00.310 --> 00:40:01.143
Thank you for your time.
902
00:40:01.143 --> 00:40:02.337
Thank you to NIDA and the NIHAAA
903
00:40:02.337 --> 00:40:04.340
for this opportunity to present my research,
904
00:40:04.340 --> 00:40:05.907
and thanks to my labmates and team
905
00:40:05.907 --> 00:40:07.460
at Rutgers and our funding support.
906
00:40:07.460 --> 00:40:08.760
If you'd like any more information,
907
00:40:08.760 --> 00:40:10.090
please see our recent publication
908
00:40:10.090 --> 00:40:12.150
in The International General Psychophysiology,
909
00:40:12.150 --> 00:40:14.350
and of course I'd be happy to answer any questions here.
910
00:40:14.350 --> 00:40:15.460
Thank you.
911
00:40:15.460 --> 00:40:16.580
Okay, thank you Kathryn.
912
00:40:16.580 --> 00:40:19.700
We got a couple questions in on our chat.
913
00:40:19.700 --> 00:40:23.993
First one is can you speculate on the mechanism of delay?
914
00:40:25.470 --> 00:40:26.840
Mechanism of delay?
915
00:40:26.840 --> 00:40:27.940
Yes.
916
00:40:27.940 --> 00:40:32.570
So what we anticipate is that we didn't actually apply TMS
917
00:40:32.570 --> 00:40:35.642
for the first 100 trials of the test
918
00:40:35.642 --> 00:40:36.644
just to determine whether or not
919
00:40:36.644 --> 00:40:38.620
there was a baseline impaired reward positivity
920
00:40:38.620 --> 00:40:40.410
across both groups.
921
00:40:40.410 --> 00:40:45.070
And then in terms of how we anticipate that the TMS
922
00:40:45.070 --> 00:40:48.900
builds up over time is that the DLPFC needs time
923
00:40:48.900 --> 00:40:52.160
to enhance the reward prediction error signals
924
00:40:52.160 --> 00:40:55.070
that are being transferred to the anterior singular cortex
925
00:40:55.070 --> 00:40:58.780
and striatum, that way sort of building up
926
00:40:58.780 --> 00:41:02.183
those positive reward prediction error signals over time.
927
00:41:03.990 --> 00:41:06.000
Okay, we got another question.
928
00:41:06.000 --> 00:41:07.230
Is there a difference
929
00:41:07.230 --> 00:41:10.290
between the drug being used in the response?
930
00:41:10.290 --> 00:41:12.524
It appears the response to opioids is greatest.
931
00:41:12.524 --> 00:41:17.260
So we did have a opioid-heavy using sample
932
00:41:17.260 --> 00:41:18.710
that was just what we have obviously we working
933
00:41:18.710 --> 00:41:22.226
on New Jersey which has a significant (indistinct) problem,
934
00:41:22.226 --> 00:41:25.190
but when we factored in different types of drugs
935
00:41:25.190 --> 00:41:26.590
using that (indistinct) factor
936
00:41:26.590 --> 00:41:28.710
then there was no difference between drug groups.
937
00:41:28.710 --> 00:41:29.543
Okay.
938
00:41:29.543 --> 00:41:32.480
And another question came in, I'm curious to know
939
00:41:32.480 --> 00:41:37.480
if participants who received real TMS reported perceiving
940
00:41:37.890 --> 00:41:42.470
an easier time completing the PST task
941
00:41:42.470 --> 00:41:45.160
compared to the SHAM subjects.
942
00:41:45.160 --> 00:41:47.720
So we didn't actually take any anecdotal evidence,
943
00:41:47.720 --> 00:41:51.640
but we sort of, from the anecdotal stuff
944
00:41:51.640 --> 00:41:53.240
that we spoke to with participants,
945
00:41:53.240 --> 00:41:54.530
they couldn't really tell the difference,
946
00:41:54.530 --> 00:41:56.210
so as long as there was a noise going on,
947
00:41:56.210 --> 00:41:58.310
everyone thought that they were saved TMS.
948
00:41:59.710 --> 00:42:02.817
Okay, and it looks like, let's see, we got a couple more,
949
00:42:02.817 --> 00:42:07.190
what TMS stimulator model was used?
950
00:42:07.190 --> 00:42:11.214
Some TMS can produce perceptible temperature changes,
951
00:42:11.214 --> 00:42:14.573
even pain at the site of the stimulation.
952
00:42:15.420 --> 00:42:18.020
We have a (indistinct) coil butterfly that we use.
953
00:42:18.020 --> 00:42:20.558
And I figured out butterfly coil.
954
00:42:20.558 --> 00:42:25.420
We didn't experience, we take a post-TMS survey to determine
955
00:42:25.420 --> 00:42:27.990
whether or not people felt any external effects.
956
00:42:27.990 --> 00:42:31.140
And we had no external effects of temperature, pain
957
00:42:31.140 --> 00:42:34.040
or anything like that reported by any of the participants.
958
00:42:35.000 --> 00:42:36.317
Okay, thank you very much.
959
00:42:36.317 --> 00:42:39.360
And we can answer more questions at the end.
960
00:42:39.360 --> 00:42:42.730
I now wanna turn the Early Career Investigators Showcase
961
00:42:42.730 --> 00:42:45.150
over to my co-chair and colleague,
962
00:42:45.150 --> 00:42:48.120
Dr. Roger Sorenson from NIDA.
963
00:42:48.120 --> 00:42:50.760
Roger, it's all yours (laughing)
964
00:42:50.760 --> 00:42:52.170
Thanks John.
965
00:42:52.170 --> 00:42:55.000
Yeah, so let's move into the second half
966
00:42:55.000 --> 00:42:57.050
of this session here.
967
00:42:57.050 --> 00:42:59.600
And so the next presenter will be Laura Ferguson
968
00:42:59.600 --> 00:43:03.113
from the Dell Medical School at the UT in Austin.
969
00:43:04.240 --> 00:43:05.850
Laura, please.
970
00:43:05.850 --> 00:43:07.920
Hi, I'm Laura Ferguson,
971
00:43:07.920 --> 00:43:10.840
and I'm a post-doc in Bob Messing's Lab
972
00:43:10.840 --> 00:43:13.620
at the Dell Medical School at UT Austin.
973
00:43:13.620 --> 00:43:15.010
And I'm very grateful
974
00:43:15.010 --> 00:43:17.050
to the selection committee for giving me
975
00:43:17.050 --> 00:43:20.500
this opportunity to present some unpublished data
976
00:43:20.500 --> 00:43:24.670
I've gathered searching for an accessible transcriptome
977
00:43:24.670 --> 00:43:28.970
to use for guiding diagnostic and treatments
978
00:43:28.970 --> 00:43:31.102
for alcohol use disorders, or AUD.
979
00:43:31.102 --> 00:43:36.010
So for the background, next side,
980
00:43:36.010 --> 00:43:37.720
when I was in graduate school
981
00:43:37.720 --> 00:43:40.570
in Adrian Harris and Dane Mayfield's labs,
982
00:43:40.570 --> 00:43:42.060
I had the opportunity to work
983
00:43:42.060 --> 00:43:46.050
on a project where we used brain gene expression profiles
984
00:43:46.050 --> 00:43:47.260
from a mouse model
985
00:43:47.260 --> 00:43:49.240
of intense binge-like
986
00:43:49.240 --> 00:43:52.140
drinking to predict treatments that ended up
987
00:43:52.140 --> 00:43:54.950
drastically reducing the amount of ethanol consumed
988
00:43:54.950 --> 00:43:58.173
and blood alcohol levels attained in those mice.
989
00:43:59.050 --> 00:44:02.410
And previous work in the Harris and Mayfield labs
990
00:44:02.410 --> 00:44:05.850
had used human post-mortem brain tissue
991
00:44:05.850 --> 00:44:09.660
to use gene expression profiles that were able
992
00:44:09.660 --> 00:44:11.480
to discriminate between alcoholic
993
00:44:11.480 --> 00:44:13.950
and non-alcoholic individuals.
994
00:44:13.950 --> 00:44:16.830
But the problem is we don't have access
995
00:44:16.830 --> 00:44:19.780
to brain tissue from living patients.
996
00:44:19.780 --> 00:44:24.020
So if we would like to use gene expression profiling
997
00:44:24.020 --> 00:44:27.610
to guide diagnostic or treatment decisions for AUD,
998
00:44:27.610 --> 00:44:30.960
we need to find a more accessible transcriptome.
999
00:44:30.960 --> 00:44:32.223
Next side, please.
1000
00:44:33.140 --> 00:44:35.610
So whole blood is an accessible tissue
1001
00:44:35.610 --> 00:44:39.380
and has been shown to have disease-relevant information
1002
00:44:39.380 --> 00:44:42.623
for other brain diseases, such as schizophrenia and PTSD,
1003
00:44:42.623 --> 00:44:45.240
just to name a couple.
1004
00:44:45.240 --> 00:44:47.230
And so that led us to hypothesize
1005
00:44:47.230 --> 00:44:50.520
that the blood transcriptome may share enough features
1006
00:44:50.520 --> 00:44:53.537
with the brain to be useful in diagnosing AUD.
1007
00:44:54.410 --> 00:44:57.980
And specifically, we wanted to address these two questions,
1008
00:44:57.980 --> 00:45:01.600
first, how similar our blood and brain transcriptomes?
1009
00:45:01.600 --> 00:45:04.450
And this necessitates the use of an animal model
1010
00:45:04.450 --> 00:45:06.630
so that you can compare the brain
1011
00:45:06.630 --> 00:45:09.360
and blood from the same subject.
1012
00:45:09.360 --> 00:45:10.610
And the second question
1013
00:45:10.610 --> 00:45:14.470
is can blood gene expression profiles be used
1014
00:45:14.470 --> 00:45:18.870
to predict alcohol dependent and non-dependent subjects?
1015
00:45:18.870 --> 00:45:20.053
Next side, please.
1016
00:45:21.090 --> 00:45:23.030
So this is the experimental design.
1017
00:45:23.030 --> 00:45:25.740
And I won't go into detail except to say
1018
00:45:25.740 --> 00:45:29.340
that we used the chronic intermittent ethanol procedure
1019
00:45:30.260 --> 00:45:34.440
where mice are exposed passively to high amounts of ethanol.
1020
00:45:34.440 --> 00:45:38.420
And that was interspersed with voluntary drinking test
1021
00:45:38.420 --> 00:45:41.000
in two-bottle choice drinking tests.
1022
00:45:41.000 --> 00:45:44.170
So we used this procedure to induce alcohol dependence
1023
00:45:44.170 --> 00:45:48.440
in the mice, and the control group only had air exposure
1024
00:45:48.440 --> 00:45:52.190
in the vapor chambers, but were allowed to consume ethanol
1025
00:45:52.190 --> 00:45:54.630
in the voluntary ethanol tests.
1026
00:45:54.630 --> 00:45:57.570
And it's known that gene expression can be volatile
1027
00:45:57.570 --> 00:45:59.230
right after this procedure.
1028
00:45:59.230 --> 00:46:02.694
So we waited a week after the last vapor exposure
1029
00:46:02.694 --> 00:46:06.426
to measure gene expression in brain and whole blood to look
1030
00:46:06.426 --> 00:46:10.040
at long-term changes in gene expression
1031
00:46:10.040 --> 00:46:12.350
during alcohol withdrawal.
1032
00:46:12.350 --> 00:46:14.320
Next side, please.
1033
00:46:14.320 --> 00:46:16.090
So to get at his first question,
1034
00:46:16.090 --> 00:46:19.441
do blood gene expression levels reflect those in brain?
1035
00:46:19.441 --> 00:46:23.760
We looked at all subjects irrespective of their treatment
1036
00:46:23.760 --> 00:46:26.350
and calculated the correlation coefficient
1037
00:46:26.350 --> 00:46:28.960
between blood gene expression levels
1038
00:46:30.006 --> 00:46:30.839
and brain gene expression levels
1039
00:46:30.839 --> 00:46:32.290
for these three brain areas.
1040
00:46:32.290 --> 00:46:34.330
And you can see that there were hundreds
1041
00:46:34.330 --> 00:46:36.510
of genes even after correcting
1042
00:46:36.510 --> 00:46:39.440
for multiple comparisons that were correlated
1043
00:46:39.440 --> 00:46:41.450
between blood and brain.
1044
00:46:41.450 --> 00:46:42.483
And then next,
1045
00:46:44.980 --> 00:46:47.710
we wanted to address a biological question pertaining
1046
00:46:47.710 --> 00:46:48.980
to alcohol withdrawal.
1047
00:46:48.980 --> 00:46:51.560
And we wanted to see if the genes differentially expressed
1048
00:46:51.560 --> 00:46:55.090
during withdrawal were the same between blood and brain.
1049
00:46:55.090 --> 00:46:57.800
And you can see that the overlap was quite small
1050
00:46:57.800 --> 00:47:02.230
and the significance was only striking for the males,
1051
00:47:02.230 --> 00:47:04.610
prefrontal cortex and hypothalamus
1052
00:47:04.610 --> 00:47:06.920
and borderline or trending significant
1053
00:47:06.920 --> 00:47:10.020
for the female hypothalamus in the make delay.
1054
00:47:10.020 --> 00:47:13.260
But we know from human post-mortem brain tissue
1055
00:47:13.260 --> 00:47:16.700
that differences in gene expression levels between alcoholic
1056
00:47:16.700 --> 00:47:19.740
and non-alcoholic subjects is quite small
1057
00:47:19.740 --> 00:47:22.010
compared to differences in gene networks
1058
00:47:22.010 --> 00:47:25.060
or how the genes interact with one another.
1059
00:47:25.060 --> 00:47:28.730
So next we wanted to look at the gene networks.
1060
00:47:28.730 --> 00:47:29.870
And when we did that,
1061
00:47:29.870 --> 00:47:34.390
we found three main categories of gene networks perturbed
1062
00:47:34.390 --> 00:47:39.040
during an alcohol withdrawal that were overlapping
1063
00:47:39.040 --> 00:47:41.070
between blood and brain.
1064
00:47:41.070 --> 00:47:42.810
And in the first network,
1065
00:47:42.810 --> 00:47:45.540
it's number one here, the cell-to-cell signaling network,
1066
00:47:45.540 --> 00:47:47.820
the overlap was particularly strong
1067
00:47:47.820 --> 00:47:52.060
between blood hypothalamus and prefrontal cortex.
1068
00:47:52.060 --> 00:47:54.730
And the other two networks that were shared
1069
00:47:54.730 --> 00:47:58.500
between blood and brain were related to immune responses
1070
00:47:58.500 --> 00:48:02.193
and protein processing and mitochondrial function.
1071
00:48:04.340 --> 00:48:05.300
Next side, please.
1072
00:48:05.300 --> 00:48:07.330
And to get at the second question,
1073
00:48:07.330 --> 00:48:10.690
can blood gene expression predict alcohol dependent
1074
00:48:10.690 --> 00:48:12.600
and non-dependent subjects?
1075
00:48:12.600 --> 00:48:15.430
We use the blood gene expression profiles
1076
00:48:15.430 --> 00:48:17.900
to train machine learning classifiers.
1077
00:48:17.900 --> 00:48:21.610
And I'm only showing two here for simplicity's sake.
1078
00:48:21.610 --> 00:48:25.550
And when our input were the differentially-expressed genes,
1079
00:48:25.550 --> 00:48:28.613
the classification accuracy was perfect or near perfect.
1080
00:48:28.613 --> 00:48:32.550
And the other inputs were less accurate
1081
00:48:32.550 --> 00:48:34.750
at classifying the subjects.
1082
00:48:34.750 --> 00:48:36.760
However, all performed better
1083
00:48:36.760 --> 00:48:39.610
than when randomly selecting genes.
1084
00:48:39.610 --> 00:48:43.040
And this tells us that there is information
1085
00:48:43.040 --> 00:48:46.920
in the whole blood transcriptome related
1086
00:48:46.920 --> 00:48:50.623
to alcohol dependence even a week into withdrawal.
1087
00:48:51.870 --> 00:48:53.600
Next slide please.
1088
00:48:53.600 --> 00:48:55.460
So to summarize what I've shown you,
1089
00:48:55.460 --> 00:48:56.990
the answer to the first question,
1090
00:48:56.990 --> 00:48:59.990
do blood gene expression patterns reflect those in brain?
1091
00:48:59.990 --> 00:49:01.210
The answer is partially.
1092
00:49:01.210 --> 00:49:02.060
So when we looked
1093
00:49:02.060 --> 00:49:04.700
at the population level and included all subjects
1094
00:49:04.700 --> 00:49:07.380
irrespective of their treatment, we found hundreds
1095
00:49:07.380 --> 00:49:10.800
of genes correlated between blood and brain.
1096
00:49:10.800 --> 00:49:15.030
There was a small overlap of the specific genes
1097
00:49:15.030 --> 00:49:17.130
differentially expressed during withdrawal.
1098
00:49:17.130 --> 00:49:18.870
However, there was considerable overlap
1099
00:49:18.870 --> 00:49:21.570
between the gene networks perturbed during withdrawal.
1100
00:49:22.690 --> 00:49:24.390
And the answer to the second question,
1101
00:49:24.390 --> 00:49:27.620
can blood gene expression profiles predict alcohol dependent
1102
00:49:27.620 --> 00:49:29.510
and non-dependent subjects?
1103
00:49:29.510 --> 00:49:33.490
That answer seems to be a resounding yes.
1104
00:49:33.490 --> 00:49:36.090
And however promising these results are,
1105
00:49:36.090 --> 00:49:37.690
they do need to be validated
1106
00:49:37.690 --> 00:49:40.870
in larger and more heterogeneous populations.
1107
00:49:40.870 --> 00:49:44.710
And we do plan on looking
1108
00:49:44.710 --> 00:49:48.040
at human blood gene expression data sets to see
1109
00:49:48.040 --> 00:49:50.690
if we can get similar results there.
1110
00:49:50.690 --> 00:49:53.340
And with that, I'd like to conclude my talk
1111
00:49:53.340 --> 00:49:56.380
by acknowledging the people who have contributed
1112
00:49:56.380 --> 00:49:59.210
to this work and my funding sources.
1113
00:49:59.210 --> 00:50:02.310
And I thank you for your attention and time,
1114
00:50:02.310 --> 00:50:04.053
and I'm happy to answer questions.
1115
00:50:05.213 --> 00:50:07.100
Thanks much (clears throat)
1116
00:50:07.100 --> 00:50:08.660
Thanks much, Laura.
1117
00:50:08.660 --> 00:50:13.280
So our first question we have here is what were the BACs
1118
00:50:13.280 --> 00:50:15.435
of the mice used?
1119
00:50:15.435 --> 00:50:20.435
They ranged from 175 maybe to 200 milligrams percent.
1120
00:50:22.542 --> 00:50:25.173
Okay.
1121
00:50:26.229 --> 00:50:30.180
Have you considered testing the ideas
1122
00:50:30.180 --> 00:50:32.810
in other strains of mice?
1123
00:50:32.810 --> 00:50:33.643
Yes.
1124
00:50:33.643 --> 00:50:35.350
That would be great to test,
1125
00:50:35.350 --> 00:50:38.160
especially in like some out-bred mice
1126
00:50:38.160 --> 00:50:40.270
and heterogeneous populations.
1127
00:50:40.270 --> 00:50:42.630
We started with the C57 mice
1128
00:50:42.630 --> 00:50:45.620
mostly because we wanted to be able to compare
1129
00:50:45.620 --> 00:50:48.790
the brain transcriptome to previous studies
1130
00:50:48.790 --> 00:50:49.630
that have been conducted
1131
00:50:49.630 --> 00:50:52.849
which were done in this C57 black six.
1132
00:50:52.849 --> 00:50:53.851
Mm-hmm.
1133
00:50:53.851 --> 00:50:55.684
And the follow-up with that,
1134
00:50:55.684 --> 00:50:57.910
what about the (indistinct) mice?
1135
00:50:57.910 --> 00:51:00.940
Yeah, that would be great to look at that too.
1136
00:51:00.940 --> 00:51:05.390
Yeah, specially models have specific characteristics
1137
00:51:05.390 --> 00:51:09.963
and risks for AUD, would be really nice to look at that.
1138
00:51:12.680 --> 00:51:15.623
Did you consider using exosome sequencing?
1139
00:51:17.210 --> 00:51:22.210
Yeah, the idea that the blood touches every organ
1140
00:51:22.350 --> 00:51:25.930
including the brain, and perhaps exosomes would be promising
1141
00:51:25.930 --> 00:51:29.080
for passing information between blood and brain
1142
00:51:29.080 --> 00:51:30.890
is a great idea.
1143
00:51:30.890 --> 00:51:33.866
It's less, I guess we wanted to start
1144
00:51:33.866 --> 00:51:38.840
with the most clinically straightforward and simple.
1145
00:51:38.840 --> 00:51:41.590
So start simple if you find something great,
1146
00:51:41.590 --> 00:51:44.330
if you don't get more complex.
1147
00:51:44.330 --> 00:51:45.500
So that was the thinking
1148
00:51:45.500 --> 00:51:47.793
behind starting with just whole blood.
1149
00:51:49.720 --> 00:51:53.890
Question here, does blood RNA include only cellular
1150
00:51:53.890 --> 00:51:55.927
or both cellular and vesicular?
1151
00:51:56.880 --> 00:52:00.410
So the cellular would be coming from the white blood cells
1152
00:52:00.410 --> 00:52:05.000
but there is some RNA in just the serum.
1153
00:52:05.000 --> 00:52:09.893
So this would include all extracellular and cellular RNA,
1154
00:52:09.893 --> 00:52:11.380
which is a benefit of using whole blood
1155
00:52:11.380 --> 00:52:14.400
instead of just like the PBMs,
1156
00:52:14.400 --> 00:52:17.009
the white blood cells, for example.
1157
00:52:17.009 --> 00:52:18.030
Yeah.
1158
00:52:18.030 --> 00:52:21.403
Yeah, and as far as the classes that you pointed out,
1159
00:52:22.793 --> 00:52:26.400
can you further define the classes?
1160
00:52:26.400 --> 00:52:30.580
Are, are the genes related to more excitability changes
1161
00:52:30.580 --> 00:52:35.580
or degeneration or something like that?
1162
00:52:36.330 --> 00:52:38.660
Yeah, so the cell-to-cell signaling,
1163
00:52:38.660 --> 00:52:41.220
that was just a very broad category
1164
00:52:41.220 --> 00:52:45.550
I gave to a lot of biological pathways,
1165
00:52:45.550 --> 00:52:49.870
and that included both GABA and glutamatergic signaling.
1166
00:52:49.870 --> 00:52:53.740
It included endocannabinoid signaling synaptogenesis.
1167
00:52:53.740 --> 00:52:56.480
So traditionally brain, a lot
1168
00:52:56.480 --> 00:53:00.500
of brain and neurotransmitter-type pathways
1169
00:53:00.500 --> 00:53:02.950
which it was really interesting to find that so strong
1170
00:53:02.950 --> 00:53:04.653
in the blood as well.
1171
00:53:05.862 --> 00:53:07.255
Interesting.
1172
00:53:07.255 --> 00:53:09.380
Okay, thank you very much, Laura.
1173
00:53:09.380 --> 00:53:10.870
Thank you.
1174
00:53:10.870 --> 00:53:14.050
We will move to our next presenter,
1175
00:53:14.050 --> 00:53:18.700
and pardon me 'cause I'm gonna boost your name (laughing)
1176
00:53:18.700 --> 00:53:23.700
But Mehdi Farokhnia of our Intramural Program at the NIH.
1177
00:53:25.830 --> 00:53:27.304
Thank you, and thank you so much
1178
00:53:27.304 --> 00:53:29.660
to the selection committee
1179
00:53:29.660 --> 00:53:32.610
for giving me the opportunity to present.
1180
00:53:32.610 --> 00:53:34.280
Yeah, my name is Mehdi.
1181
00:53:34.280 --> 00:53:37.750
I'm a Staff Scientist in Lorenzo Leggio Lab
1182
00:53:37.750 --> 00:53:40.080
that you heard this talk yesterday.
1183
00:53:40.080 --> 00:53:44.566
And today I'm gonna talk about one of the recent targets,
1184
00:53:44.566 --> 00:53:48.240
the mineralocorticoid receptor that we have started
1185
00:53:48.240 --> 00:53:52.120
to look at as a potential pharmacological target
1186
00:53:52.120 --> 00:53:54.590
to treat alcohol use disorder.
1187
00:53:54.590 --> 00:53:55.423
Next.
1188
00:53:56.620 --> 00:53:59.540
I don't have anything to disclose.
1189
00:53:59.540 --> 00:54:00.830
Next, please.
1190
00:54:00.830 --> 00:54:04.640
So a very brief background, the mineralocorticoid receptors
1191
00:54:04.640 --> 00:54:08.520
are part of the (indistinct) angiotensin aldosterone system,
1192
00:54:08.520 --> 00:54:12.340
and their main endogenous ligand is the hormone aldosterone,
1193
00:54:12.340 --> 00:54:15.540
and their primary function is to control fluids
1194
00:54:15.540 --> 00:54:18.850
and electrolyte homeostasis.
1195
00:54:18.850 --> 00:54:21.010
They are widely expressed in the body
1196
00:54:21.010 --> 00:54:24.220
and including the brain and the brain regions
1197
00:54:24.220 --> 00:54:26.280
that are involved in addictive behaviors
1198
00:54:26.280 --> 00:54:29.920
such as amygdala, hippocampus and PFC
1199
00:54:29.920 --> 00:54:30.753
Next.
1200
00:54:32.750 --> 00:54:37.750
So some recent data suggest that the (indistinct)
1201
00:54:38.152 --> 00:54:40.850
mineralocorticoid receptor pathway might be involved
1202
00:54:40.850 --> 00:54:42.898
in alcohol-seeking behavior.
1203
00:54:42.898 --> 00:54:44.853
I won't go into all the details.
1204
00:54:45.744 --> 00:54:47.550
This is a pretty busier slide, but our group
1205
00:54:47.550 --> 00:54:51.560
recently showed that overall higher functional activity
1206
00:54:51.560 --> 00:54:55.900
in the mineralocorticoid endocrine pathway may contribute
1207
00:54:55.900 --> 00:54:59.270
to susceptibility to alcohol use.
1208
00:54:59.270 --> 00:55:02.690
And these data were shown across three species, humans,
1209
00:55:02.690 --> 00:55:05.150
monkeys and rats, and consistent
1210
00:55:05.150 --> 00:55:07.830
with that two recent studies
1211
00:55:07.830 --> 00:55:11.350
from Joyce Besheer Lab at UNC has shown
1212
00:55:11.350 --> 00:55:13.699
that MR antagonism
1213
00:55:13.699 --> 00:55:17.240
seemed to reduce alcohol self-administration.
1214
00:55:17.240 --> 00:55:19.240
So this has led us to ask the question
1215
00:55:19.240 --> 00:55:22.530
whether a mineralocorticoid receptor
1216
00:55:22.530 --> 00:55:24.620
could be a pharmacotherapeutic target
1217
00:55:24.620 --> 00:55:26.103
for alcohol use disorder.
1218
00:55:27.069 --> 00:55:28.900
And fortunately, to test this hypothesis,
1219
00:55:28.900 --> 00:55:32.393
we do have a readily available tool, an spironolactone,
1220
00:55:33.310 --> 00:55:35.680
which is an MR antagonist FDA approved
1221
00:55:36.786 --> 00:55:38.310
for different indications, and it's widely used
1222
00:55:38.310 --> 00:55:42.570
in primary care, mostly to treat hypertension and edema
1223
00:55:43.670 --> 00:55:44.503
Next.
1224
00:55:45.350 --> 00:55:47.390
So in the first study in collaboration
1225
00:55:47.390 --> 00:55:50.020
with Dr. Koob Lab at NIDA, we tested the effects
1226
00:55:50.020 --> 00:55:53.540
of spironolactone on alcohol self-administration
1227
00:55:53.540 --> 00:55:57.810
in male (indistinct) rats who were made dependent on alcohol
1228
00:55:57.810 --> 00:55:59.630
by chronic alcohol vapor exposure,
1229
00:55:59.630 --> 00:56:01.660
and also non-dependent rats.
1230
00:56:01.660 --> 00:56:05.260
Again, I don't have time to go into all details, but overall
1231
00:56:05.260 --> 00:56:07.590
as you see on the figure on the left,
1232
00:56:07.590 --> 00:56:10.590
both in alcohol-dependent and non-dependent rats,
1233
00:56:10.590 --> 00:56:14.770
the spironolactone reduced alcohol self-administration,
1234
00:56:14.770 --> 00:56:18.530
and these data were also replicated in a separate cohort
1235
00:56:18.530 --> 00:56:21.690
as you see on the figure on the right, with the highest dose
1236
00:56:21.690 --> 00:56:25.000
that both independent and non-dependent rats,
1237
00:56:25.000 --> 00:56:26.250
basically the spironolactone
1238
00:56:26.250 --> 00:56:28.830
reduced alcohol self-administration.
1239
00:56:28.830 --> 00:56:29.663
Next
1240
00:56:30.940 --> 00:56:33.107
In the second study,
1241
00:56:33.107 --> 00:56:35.210
again, in collaboration with Dr. Koob's Lab,
1242
00:56:35.210 --> 00:56:36.530
we tested the spironolactone
1243
00:56:36.530 --> 00:56:40.210
in a different model in the drinking in the dark paradigm
1244
00:56:40.210 --> 00:56:44.080
which is a well-validated model of alcohol binge-drinking.
1245
00:56:44.080 --> 00:56:47.370
And here we used both male and female mice.
1246
00:56:47.370 --> 00:56:51.430
And again, as you see on both figures, both the male
1247
00:56:53.339 --> 00:56:55.280
and females, spironolactone dose dependently reduced
1248
00:56:55.280 --> 00:56:58.235
alcohol self-administration in this model
1249
00:56:58.235 --> 00:56:59.453
Next.
1250
00:57:01.070 --> 00:57:04.600
So in parallel to this preclinical data, as I mentioned,
1251
00:57:04.600 --> 00:57:08.110
the spironolactone is FDA approved for, so it does have
1252
00:57:08.110 --> 00:57:11.100
a promise for drug repurposing, and this elegant review
1253
00:57:11.100 --> 00:57:13.680
recently summarizes different ways
1254
00:57:13.680 --> 00:57:16.220
of approaching drug repurposing.
1255
00:57:16.220 --> 00:57:20.710
And we have been doing some pharmaco epi studies looking
1256
00:57:20.710 --> 00:57:24.150
at electronic health records data, trying to see
1257
00:57:24.150 --> 00:57:27.651
if we can get a signal in humans, and if the signal
1258
00:57:27.651 --> 00:57:28.865
is strong enough to move to a clinical trial.
1259
00:57:28.865 --> 00:57:33.865
So we are looking at the effect of the spironolactone
1260
00:57:34.290 --> 00:57:37.940
prescribed for any indication on alcohol use.
1261
00:57:37.940 --> 00:57:38.773
Next.
1262
00:57:42.600 --> 00:57:45.560
So for the first Pharmaco epi study, we collaborated
1263
00:57:45.560 --> 00:57:48.620
with Kaiser Permanente in North California.
1264
00:57:48.620 --> 00:57:51.550
And as you know, this is an integrated healthcare system
1265
00:57:51.550 --> 00:57:54.200
which serves around 5 million members.
1266
00:57:54.200 --> 00:57:58.290
And in 2013, they actually implemented a new system
1267
00:57:58.290 --> 00:58:01.911
where they are evaluating alcohol drinking
1268
00:58:01.911 --> 00:58:04.400
as a vital sign.
1269
00:58:04.400 --> 00:58:07.480
As you can see here on this snapshot, for example,
1270
00:58:07.480 --> 00:58:10.650
your healthcare providers there, they get triggered
1271
00:58:10.650 --> 00:58:13.585
to ask alcohol-related questions as part of the screening
1272
00:58:13.585 --> 00:58:17.520
and similar to other vital signs.
1273
00:58:17.520 --> 00:58:21.145
And they collect data on daily alcohol use,
1274
00:58:21.145 --> 00:58:24.910
and also they can calculate the drinks per week
1275
00:58:24.910 --> 00:58:25.743
Next.
1276
00:58:28.350 --> 00:58:30.720
So again, our question here was to see
1277
00:58:30.720 --> 00:58:33.220
whether a spironolactone prescription is associated
1278
00:58:33.220 --> 00:58:35.210
with reduced alcohol consumption.
1279
00:58:35.210 --> 00:58:36.360
And for this, we used
1280
00:58:36.360 --> 00:58:39.360
high-dimensional propensity score matching.
1281
00:58:39.360 --> 00:58:42.128
Very briefly, we have an inception period here,
1282
00:58:42.128 --> 00:58:46.510
and treated individuals are those who get spironolactone
1283
00:58:46.510 --> 00:58:48.690
during the inception period.
1284
00:58:48.690 --> 00:58:53.150
We have a baseline and follow-up, and basically
1285
00:58:53.150 --> 00:58:55.050
each treated individual is matched
1286
00:58:55.050 --> 00:58:58.400
to five untreated individuals based on a long list
1287
00:58:58.400 --> 00:59:00.310
of co-variates, and the goal is to see
1288
00:59:00.310 --> 00:59:04.390
whether spironolactone used is associated with change
1289
00:59:04.390 --> 00:59:07.510
in alcohol consumption from baseline to follow-up
1290
00:59:07.510 --> 00:59:08.343
Next.
1291
00:59:09.720 --> 00:59:11.400
This is just to show the numbers.
1292
00:59:11.400 --> 00:59:14.000
And so in this data set,
1293
00:59:14.000 --> 00:59:17.550
we had around 40,000 treated subjects with spironolactone
1294
00:59:17.550 --> 00:59:19.670
and 5 million people untreated.
1295
00:59:19.670 --> 00:59:21.420
As you see, we have a very list
1296
00:59:21.420 --> 00:59:24.850
of inclusion-exclusion criteria to make sure that the data
1297
00:59:24.850 --> 00:59:27.870
is as clean as possible and the samples are balanced.
1298
00:59:27.870 --> 00:59:30.210
And we ended up with around 500 subjects
1299
00:59:30.210 --> 00:59:33.290
in the treated group and around 2,000 subjects
1300
00:59:33.290 --> 00:59:34.560
in the untreated group.
1301
00:59:34.560 --> 00:59:35.393
Next.
1302
00:59:36.740 --> 00:59:37.770
These are the results.
1303
00:59:37.770 --> 00:59:42.570
They are all unpublished as you can see,
1304
00:59:42.570 --> 00:59:46.510
spironolactone use compared to non-spironolactone use
1305
00:59:46.510 --> 00:59:48.520
was associated with a greater reduction
1306
00:59:48.520 --> 00:59:52.250
of drinks per week from baseline to follow-up.
1307
00:59:52.250 --> 00:59:54.740
And I'm not showing the data here, but if you want
1308
00:59:54.740 --> 00:59:58.140
you can see my poster that based on alcohol use moderated
1309
00:59:58.140 --> 01:00:01.520
this association, essentially the effect was much stronger
1310
01:00:01.520 --> 01:00:04.930
in those who had higher alcohol use at baseline.
1311
01:00:04.930 --> 01:00:07.453
And yeah, next please.
1312
01:00:09.050 --> 01:00:12.420
And most importantly probably we also founded
1313
01:00:12.420 --> 01:00:13.930
a dose response.
1314
01:00:13.930 --> 01:00:16.170
So again, as you can see here,
1315
01:00:16.170 --> 01:00:17.830
there was a significant dose response
1316
01:00:17.830 --> 01:00:20.570
of spironolactone use on reduced alcohol use.
1317
01:00:20.570 --> 01:00:23.650
Basically each 10 milligram of the spironolactone
1318
01:00:23.650 --> 01:00:28.040
was associated with a reduction of 0.36 drinks per week,
1319
01:00:28.040 --> 01:00:32.733
kind of providing some evidence for causality.
1320
01:00:33.840 --> 01:00:35.420
Next.
1321
01:00:35.420 --> 01:00:38.630
And very briefly, we, we replicated these data.
1322
01:00:38.630 --> 01:00:40.650
I'll go through this very quickly,
1323
01:00:40.650 --> 01:00:43.060
in the U.S. veteran birth cohort study,
1324
01:00:43.060 --> 01:00:44.540
this has a different design
1325
01:00:44.540 --> 01:00:48.310
of it's an prospective observational cohort study.
1326
01:00:48.310 --> 01:00:50.190
You can see the numbers here.
1327
01:00:50.190 --> 01:00:54.390
And here we looked at the different outcome, AUDIT-C score.
1328
01:00:54.390 --> 01:00:56.720
And again, a spironolactone was associated
1329
01:00:56.720 --> 01:00:58.150
with reduction alcohol use
1330
01:00:58.150 --> 01:01:00.360
in both AUD and non AUD subjects.
1331
01:01:00.360 --> 01:01:01.193
Next, please.
1332
01:01:03.560 --> 01:01:06.790
And again, the highest, basically largest effects
1333
01:01:08.360 --> 01:01:09.830
were observed among those who had higher severity
1334
01:01:09.830 --> 01:01:14.390
of alcohol use at baseline and received the highest dose.
1335
01:01:14.390 --> 01:01:15.223
Next.
1336
01:01:16.240 --> 01:01:17.640
So just to summarize,
1337
01:01:17.640 --> 01:01:20.590
we showed that the spironolactone administration
1338
01:01:20.590 --> 01:01:23.970
in rodents reduced alcohol consumption in humans.
1339
01:01:23.970 --> 01:01:26.160
Spironolactone use was associated
1340
01:01:26.160 --> 01:01:27.810
with reduced alcohol consumption.
1341
01:01:28.649 --> 01:01:31.800
And also it seems that MR antagonism could be considered
1342
01:01:31.800 --> 01:01:33.660
as a potential pharmacotherapy.
1343
01:01:33.660 --> 01:01:36.040
And hopefully if the data keeps to be promising,
1344
01:01:36.040 --> 01:01:40.220
we basically plan to do a randomized clinical trial.
1345
01:01:40.220 --> 01:01:41.053
Next, please.
1346
01:01:41.980 --> 01:01:44.520
Yeah, finally very briefly, just wanted to thank all people
1347
01:01:44.520 --> 01:01:46.747
in the lab, especially Lorenzo Leggio
1348
01:01:46.747 --> 01:01:51.030
who has been the best mentor/boss/friend (laughing)
1349
01:01:51.030 --> 01:01:52.500
that I could ask for.
1350
01:01:52.500 --> 01:01:55.430
And also all of our collaborators at the NIH IRP
1351
01:01:56.280 --> 01:01:57.890
and Yale and Kaiser Permanente
1352
01:01:57.890 --> 01:01:59.190
that they showed the data.
1353
01:02:00.174 --> 01:02:01.673
With that, I'll take any questions.
1354
01:02:02.541 --> 01:02:03.374
Thank you very much.
1355
01:02:03.374 --> 01:02:04.630
Thank you.
1356
01:02:04.630 --> 01:02:06.873
I may have missed it, was it dosing daily?
1357
01:02:08.020 --> 01:02:08.853
Yes.
1358
01:02:08.853 --> 01:02:11.060
Yeah, the doses that I showed was daily.
1359
01:02:11.060 --> 01:02:12.420
Yeah, okay.
1360
01:02:12.420 --> 01:02:14.660
We have a question here (coughs)
1361
01:02:14.660 --> 01:02:17.730
Given glucocorticoids also have high affinity
1362
01:02:17.730 --> 01:02:21.340
for the mineral receptor, how do you know
1363
01:02:21.340 --> 01:02:23.020
that you're blocking the action
1364
01:02:23.020 --> 01:02:27.833
of aldosterone and not corticosterone in your model?
1365
01:02:28.870 --> 01:02:30.400
Yeah, that's a very good point.
1366
01:02:30.400 --> 01:02:32.620
And the question, the answer is we don't know.
1367
01:02:32.620 --> 01:02:36.420
These are again, very, very preliminary data.
1368
01:02:36.420 --> 01:02:38.830
We have started testing the spironolactone
1369
01:02:38.830 --> 01:02:43.300
because it's a generic drug widely using primary care,
1370
01:02:43.300 --> 01:02:45.190
but what we are doing now,
1371
01:02:45.190 --> 01:02:47.800
again with Dr. Koob's Lab at NIDA
1372
01:02:47.800 --> 01:02:52.800
and (indistinct) is to test more MR-specific drugs,
1373
01:02:53.700 --> 01:02:57.972
specifically eplerenone which is more specific for MR
1374
01:02:57.972 --> 01:03:01.320
to see if the effects that we are seeing are really MR
1375
01:03:01.320 --> 01:03:03.523
and not GR-mediated.
1376
01:03:05.285 --> 01:03:08.340
Your study design does not roll out
1377
01:03:08.340 --> 01:03:11.490
the confounding persons prescribed spironolactone
1378
01:03:13.060 --> 01:03:15.270
may have had worsening disease
1379
01:03:15.270 --> 01:03:18.973
and greater emphasis on need to reduce alcohol consumption.
1380
01:03:20.130 --> 01:03:21.813
Is that a statement or a question?
1381
01:03:21.813 --> 01:03:24.920
Yeah, I would say both yes and no.
1382
01:03:24.920 --> 01:03:27.250
Obviously this is not a clinical trial,
1383
01:03:27.250 --> 01:03:29.520
so they are not randomized
1384
01:03:29.520 --> 01:03:32.850
and basically it's observational data.
1385
01:03:32.850 --> 01:03:35.120
But on the other hand, again, I didn't have time to go
1386
01:03:35.120 --> 01:03:39.970
into all details, but both groups are matched
1387
01:03:39.970 --> 01:03:44.970
on both predefined and empirical co-variates.
1388
01:03:45.500 --> 01:03:49.000
So for example, any visit during the last year
1389
01:03:49.000 --> 01:03:50.960
or any medication that they are taking,
1390
01:03:50.960 --> 01:03:54.610
all of the comorbidities, they are matched based on that.
1391
01:03:54.610 --> 01:03:57.240
So we are pretty confident that the both groups
1392
01:03:57.240 --> 01:04:00.440
at least in terms of health, they are balanced.
1393
01:04:00.440 --> 01:04:04.410
And for example, in the first study that I mentioned,
1394
01:04:04.410 --> 01:04:06.330
the two groups were only not balanced
1395
01:04:06.330 --> 01:04:11.330
on three factors, gender, heart failure and hyperkalemia,
1396
01:04:11.470 --> 01:04:14.250
and those three factors are controlled
1397
01:04:14.250 --> 01:04:16.010
for in all the analysis.
1398
01:04:16.010 --> 01:04:18.510
But yeah, I do agree that basically
1399
01:04:18.510 --> 01:04:21.780
we cannot rule out and to test the hypothesis.
1400
01:04:21.780 --> 01:04:23.463
We should run a clinical trial.
1401
01:04:24.350 --> 01:04:27.330
Thank you for interpreting my reading up the question.
1402
01:04:27.330 --> 01:04:29.913
Real, real quickly, any gender-related differences?
1403
01:04:31.800 --> 01:04:33.540
So we were not able to test
1404
01:04:33.540 --> 01:04:36.780
that in human data because we matched the samples
1405
01:04:36.780 --> 01:04:37.883
based on gender.
1406
01:04:39.184 --> 01:04:41.070
In the preclinical data,
1407
01:04:41.070 --> 01:04:43.885
at least in the mice data that they showed,
1408
01:04:43.885 --> 01:04:45.120
we didn't see a gender difference
1409
01:04:45.120 --> 01:04:47.060
in both groups' states reducing,
1410
01:04:47.060 --> 01:04:50.770
but in the previous slide that I showed
1411
01:04:50.770 --> 01:04:54.080
from Joyce Besheer at UNC, they did find
1412
01:04:54.080 --> 01:04:59.080
a gender difference, specifically, females were responding
1413
01:04:59.350 --> 01:05:01.494
better to a spironolactone.
1414
01:05:01.494 --> 01:05:02.930
So both groups
1415
01:05:02.930 --> 01:05:05.930
in their study reduced alcohol self-administration,
1416
01:05:05.930 --> 01:05:09.853
but the persistent was much higher in females.
1417
01:05:11.020 --> 01:05:12.226
Thank you.
1418
01:05:12.226 --> 01:05:13.710
Thank you very much.
1419
01:05:13.710 --> 01:05:14.860
Okay, we'll go on.
1420
01:05:14.860 --> 01:05:18.050
Our next speaker is Sandra Sanchez-Roige
1421
01:05:18.050 --> 01:05:21.280
from University of California San Diego.
1422
01:05:21.280 --> 01:05:22.133
Sandra.
1423
01:05:23.500 --> 01:05:24.760
I am honored to be part
1424
01:05:24.760 --> 01:05:27.490
of this session and I thank the organizers.
1425
01:05:27.490 --> 01:05:30.230
Today I will present our most recent study looking
1426
01:05:30.230 --> 01:05:32.150
into the genetic architecture
1427
01:05:32.150 --> 01:05:34.073
of alcohol use and misuse.
1428
01:05:35.120 --> 01:05:38.900
In 2018, the PTC Substance Users or the work group
1429
01:05:38.900 --> 01:05:40.980
performed the GWAS of alcohol dependence
1430
01:05:40.980 --> 01:05:43.632
which robustly implicated the world's study
1431
01:05:43.632 --> 01:05:47.690
metabolizing enzymes, gene, ADH1B and chromosome four.
1432
01:05:47.690 --> 01:05:49.300
This GWAS, which used
1433
01:05:49.300 --> 01:05:52.470
a clinically-ascertained disease population has served
1434
01:05:52.470 --> 01:05:53.743
as a gold standard.
1435
01:05:54.750 --> 01:05:58.130
Complimentary design is the use of intermediate phenotypes
1436
01:05:58.130 --> 01:06:00.560
which allows us to break down aspects
1437
01:06:00.560 --> 01:06:03.400
of alcohol and substance use disorders more generally
1438
01:06:03.400 --> 01:06:06.520
into specific components or transitions.
1439
01:06:06.520 --> 01:06:09.570
And this is important because alcohol use disorders develop
1440
01:06:09.570 --> 01:06:10.940
in a chronological fashion
1441
01:06:10.940 --> 01:06:14.610
from experimental to regular use, dependence, relapse,
1442
01:06:14.610 --> 01:06:17.960
and it is possible that different genes may impact
1443
01:06:17.960 --> 01:06:20.453
or have a different role at different stages.
1444
01:06:21.400 --> 01:06:25.110
An example of such an intermediate phenotype is AUDIT,
1445
01:06:25.110 --> 01:06:28.930
a very ten-item item questionnaire that its beauty
1446
01:06:28.930 --> 01:06:33.000
is that it can be measured in the general population.
1447
01:06:33.000 --> 01:06:34.590
Next slide, please.
1448
01:06:34.590 --> 01:06:37.743
Their biomass in very large sample sizes.
1449
01:06:40.620 --> 01:06:44.100
Its beauty is that it can also dissect alcohol use
1450
01:06:44.100 --> 01:06:47.750
from misuse, and for example, we can see in the next slides
1451
01:06:47.750 --> 01:06:51.880
that the first three items capture aspects of consumption
1452
01:06:51.880 --> 01:06:55.200
which we named AUDIT-C, and the remaining items
1453
01:06:55.200 --> 01:06:58.500
measure aspects of problematic consequences of alcohol use,
1454
01:06:58.500 --> 01:07:00.113
which we named AUDIT-P.
1455
01:07:01.050 --> 01:07:03.850
And back in 2019 with Abraham Palmer
1456
01:07:03.850 --> 01:07:06.780
and Tony Clark, we performed two independent GWAS
1457
01:07:06.780 --> 01:07:10.670
of AUDIT-C and AUDIT-P using UK Biobank data.
1458
01:07:10.670 --> 01:07:13.670
And if we can go back one slide, we can see
1459
01:07:13.670 --> 01:07:17.760
that this analysis reveal that some loci,
1460
01:07:17.760 --> 01:07:20.948
for example ethanol metabolizing enzyme genes
1461
01:07:20.948 --> 01:07:24.370
like in chromosome four were consistent across phenotypes,
1462
01:07:24.370 --> 01:07:27.740
that the overlap was incomplete suggesting that there seems
1463
01:07:27.740 --> 01:07:29.370
to be a different genetic basis
1464
01:07:29.370 --> 01:07:31.863
between alcohol use phenotypes.
1465
01:07:33.000 --> 01:07:36.540
We also performed a series of genetical relational analysis
1466
01:07:36.540 --> 01:07:39.680
and identified very distinct patterns of associations
1467
01:07:39.680 --> 01:07:41.760
between these two traits.
1468
01:07:41.760 --> 01:07:43.580
For example, AUDIT-C
1469
01:07:43.580 --> 01:07:45.810
was only moderately positively
1470
01:07:45.810 --> 01:07:48.010
genetically associated with alcohol dependence
1471
01:07:48.010 --> 01:07:51.610
as measured by the PGC, whereas AUDIT-P showed
1472
01:07:51.610 --> 01:07:55.133
stronger positive associations with the very same traits.
1473
01:07:56.200 --> 01:07:58.750
AUDIT-C was also surprisingly,
1474
01:07:58.750 --> 01:08:01.810
negatively genetically associated with some forms
1475
01:08:01.810 --> 01:08:04.620
of psychopathology and health-related outcomes
1476
01:08:04.620 --> 01:08:07.629
and positively associated with socioeconomic variables
1477
01:08:07.629 --> 01:08:10.980
whereas these associations were not observed for AUDIT-P.
1478
01:08:10.980 --> 01:08:15.360
So these were associations that puzzled many of us.
1479
01:08:15.360 --> 01:08:18.860
So these may lead us to conclude that alcohol consumption
1480
01:08:18.860 --> 01:08:21.370
measured in population-based cohorts
1481
01:08:21.370 --> 01:08:25.003
is not a good genetic proxy of alcohol use disorders.
1482
01:08:26.060 --> 01:08:27.730
Another possible explanation
1483
01:08:27.730 --> 01:08:31.370
for the discrepancies we hypothesized could be attributed
1484
01:08:31.370 --> 01:08:34.960
to genetic heterogeneity among the individual items
1485
01:08:34.960 --> 01:08:37.280
or due to the fact that these phenotypes
1486
01:08:37.280 --> 01:08:41.530
were constructed using a sum score approach that assumes
1487
01:08:41.530 --> 01:08:43.470
that each item is equally informative
1488
01:08:43.470 --> 01:08:45.680
of the construct that is being measured.
1489
01:08:45.680 --> 01:08:49.870
So in this sense, using uncorrected, if you will, weight,
1490
01:08:49.870 --> 01:08:51.490
can potentially introduce errors
1491
01:08:51.490 --> 01:08:53.160
in downstream genetic analysis,
1492
01:08:53.160 --> 01:08:54.860
such as the ones that we observed.
1493
01:08:56.020 --> 01:08:59.350
So With Travis Meyer, we used a multi-variate framework
1494
01:08:59.350 --> 01:09:01.780
with genomic structural equation modeling
1495
01:09:01.780 --> 01:09:03.760
and studied the genetic basis of each
1496
01:09:03.760 --> 01:09:07.370
of the 10 different items from AUDIT, which was measured
1497
01:09:07.370 --> 01:09:10.160
in three cohorts, primarily from the UK Biobank
1498
01:09:10.160 --> 01:09:12.093
that is not the disease population.
1499
01:09:12.990 --> 01:09:15.570
So we constructed two latent factors, consumption
1500
01:09:15.570 --> 01:09:17.440
and problems, and we can see
1501
01:09:17.440 --> 01:09:19.930
that even though they are genetically correlated,
1502
01:09:19.930 --> 01:09:23.880
they were best represented by distinct entities.
1503
01:09:23.880 --> 01:09:27.270
And I also want to note that item one highlighted
1504
01:09:27.270 --> 01:09:29.800
in gray are the frequency of al use alcohol use
1505
01:09:29.800 --> 01:09:33.550
did not load strongly onto the consumption factor
1506
01:09:33.550 --> 01:09:37.330
and that these traits, which we named frequency residual
1507
01:09:37.330 --> 01:09:39.990
carried a high genetic residual variance
1508
01:09:39.990 --> 01:09:41.280
unlike the other items
1509
01:09:41.280 --> 01:09:43.630
as I will try to unpack in the next few slides.
1510
01:09:45.210 --> 01:09:48.300
Similar to original findings for the (indistinct)
1511
01:09:48.300 --> 01:09:50.870
we again observed that the polygenic architecture
1512
01:09:50.870 --> 01:09:53.540
of the two factors is distinct, which again
1513
01:09:53.540 --> 01:09:56.130
emphasizes the value of the constructing
1514
01:09:56.130 --> 01:09:59.630
such two core symptoms of alcohol use disorders.
1515
01:09:59.630 --> 01:10:02.390
But more importantly, we also found that consumption,
1516
01:10:02.390 --> 01:10:05.330
some problems were were strongly, very strongly indeed,
1517
01:10:05.330 --> 01:10:07.790
genetically correlated with alcohol dependence
1518
01:10:07.790 --> 01:10:09.290
and that we no longer observed
1519
01:10:09.290 --> 01:10:11.130
unexpected genetic associations
1520
01:10:11.130 --> 01:10:14.370
with consumption on health and positive economic outcomes
1521
01:10:14.370 --> 01:10:15.793
like we observed before.
1522
01:10:17.020 --> 01:10:19.950
On the contrary, it was a frequency of residual here
1523
01:10:19.950 --> 01:10:21.140
the darker gray bars
1524
01:10:21.140 --> 01:10:23.490
that was negatively genetically associated
1525
01:10:23.490 --> 01:10:26.230
with alcohol dependence and other psychiatric disorders
1526
01:10:26.230 --> 01:10:28.200
and positively genetically associated
1527
01:10:28.200 --> 01:10:31.510
with socioeconomic outcomes, suggesting that many
1528
01:10:31.510 --> 01:10:33.430
of the puzzling genetical relations
1529
01:10:33.430 --> 01:10:35.880
that we and others previously reported
1530
01:10:35.880 --> 01:10:37.570
could be potentially explained
1531
01:10:37.570 --> 01:10:39.600
by socially-stratified differences
1532
01:10:39.600 --> 01:10:42.890
in behavior rather than the variance that may be related
1533
01:10:42.890 --> 01:10:46.700
to the alcohol phenotypes of clinical interest.
1534
01:10:46.700 --> 01:10:49.260
So these findings are to my view extremely important
1535
01:10:49.260 --> 01:10:51.540
because they suggest that we could use
1536
01:10:51.540 --> 01:10:54.640
consumption phenotypes contrary to our initial observations
1537
01:10:54.640 --> 01:10:57.580
for future meta-analysis of alcohol use disorders,
1538
01:10:57.580 --> 01:11:00.290
reaching unprecedented sample sizes that both enhance
1539
01:11:00.290 --> 01:11:03.540
gene discovery as well as allow us to dissect
1540
01:11:03.540 --> 01:11:06.050
the genetic contributions that may be specific
1541
01:11:06.050 --> 01:11:08.073
to different alcohol phenotypes.
1542
01:11:09.060 --> 01:11:10.910
So in closing, I have here considered
1543
01:11:10.910 --> 01:11:13.110
additional study designs with AUDIT.
1544
01:11:13.110 --> 01:11:15.900
We have began to relatively inexpensively annotate
1545
01:11:15.900 --> 01:11:19.500
different addictions stages and show that alcohol use
1546
01:11:19.500 --> 01:11:22.690
and misuse have a different genetic basis.
1547
01:11:22.690 --> 01:11:24.930
AUDIT is strongly genetically associated
1548
01:11:24.930 --> 01:11:27.100
with alcohol use disorders, but it is much easier
1549
01:11:27.100 --> 01:11:29.280
to measure at large scale.
1550
01:11:29.280 --> 01:11:31.760
It is important to recognize that the population
1551
01:11:31.760 --> 01:11:34.440
that we use will influence our results.
1552
01:11:34.440 --> 01:11:36.960
And I have shown that specific phenotypes,
1553
01:11:36.960 --> 01:11:39.520
such as the frequency of alcohol use
1554
01:11:39.520 --> 01:11:42.490
will be more influenced than others.
1555
01:11:42.490 --> 01:11:45.187
I have though presented very encouraging new methods
1556
01:11:45.187 --> 01:11:48.180
such as with genomic structural equation modeling
1557
01:11:48.180 --> 01:11:51.220
that will allow us to correct or ameliorate some
1558
01:11:52.671 --> 01:11:54.300
of the previous biases that we encountered.
1559
01:11:54.300 --> 01:11:56.850
And I want to close by saying that an additional advantage
1560
01:11:56.850 --> 01:11:59.190
of using intermediate phenotypes is the ability
1561
01:11:59.190 --> 01:12:02.900
to more directly model these rates in experimental systems
1562
01:12:02.900 --> 01:12:05.730
which will be critical in extracting biological insights
1563
01:12:05.730 --> 01:12:07.563
from GWASs in the years to come.
1564
01:12:08.650 --> 01:12:10.980
The work that I have presented today is a result of a number
1565
01:12:10.980 --> 01:12:14.300
of people, particularly Abraham Palmer, Travis Maya,
1566
01:12:14.300 --> 01:12:16.990
to members of the PGC Substance Users or the work group.
1567
01:12:16.990 --> 01:12:18.650
I am fortunate to be surrounded
1568
01:12:18.650 --> 01:12:22.140
by such inspiring colleagues, and I hope to be able
1569
01:12:22.140 --> 01:12:26.083
to acknowledge support from NIAA and NIDA in the future.
1570
01:12:27.330 --> 01:12:28.953
Happy to take any questions
1571
01:12:31.659 --> 01:12:33.643
Thank you very much Sandra.
1572
01:12:34.560 --> 01:12:37.367
Are there any questions?
1573
01:12:37.367 --> 01:12:40.762
And one just kind of a thought,
1574
01:12:40.762 --> 01:12:44.040
does your data have implications for any type
1575
01:12:44.040 --> 01:12:47.463
of personalized treatment as such in the future?
1576
01:12:49.240 --> 01:12:53.220
What this analysis show is that we could potentially
1577
01:12:53.220 --> 01:12:57.300
calculate strong genetic scores,
1578
01:12:57.300 --> 01:13:02.053
if you will, that may be used or apply for future analysis.
1579
01:13:04.350 --> 01:13:05.557
Yeah.
1580
01:13:05.557 --> 01:13:06.390
Yeah.
1581
01:13:07.951 --> 01:13:09.513
Question here.
1582
01:13:10.350 --> 01:13:14.043
What would you expect with TLSB or PACS?
1583
01:13:16.621 --> 01:13:20.447
Could you unpack a TLFB or PACS for me please?
1584
01:13:22.290 --> 01:13:26.273
If the individual would explain what that is.
1585
01:13:28.260 --> 01:13:29.920
I'm not sure.
1586
01:13:29.920 --> 01:13:31.360
Yeah.
1587
01:13:31.360 --> 01:13:34.967
Any part of it is a timeline follow-back measure?
1588
01:13:34.967 --> 01:13:35.810
Ah!
1589
01:13:35.810 --> 01:13:37.608
Oh, so different types
1590
01:13:37.608 --> 01:13:40.737
of measures that have individual-item level?
1591
01:13:40.737 --> 01:13:41.570
Yeah.
Is that the question?
1592
01:13:41.570 --> 01:13:42.920
Yeah.
Yeah, we would love,
1593
01:13:43.790 --> 01:13:46.440
the thing with genetic analysis
1594
01:13:46.440 --> 01:13:51.440
is that we need large cohorts to have sufficient power.
1595
01:13:52.400 --> 01:13:56.480
And luckily AUDIT is available at large scale.
1596
01:13:56.480 --> 01:14:01.190
And in fact we hope to perform our larger GWAS of AUDIT
1597
01:14:02.110 --> 01:14:06.090
over this year or next year and AUDIT is available
1598
01:14:06.090 --> 01:14:07.560
as we have seen in the previous talk,
1599
01:14:07.560 --> 01:14:12.316
in health systems and all the cohorts we hope to assemble.
1600
01:14:12.316 --> 01:14:15.390
So this is why we focused on AUDIT,
1601
01:14:15.390 --> 01:14:17.770
but there are other measures that surely
1602
01:14:17.770 --> 01:14:19.220
are also very informative
1603
01:14:19.220 --> 01:14:21.883
that would be great to look in the future.
1604
01:14:24.580 --> 01:14:25.626
Okay.
1605
01:14:25.626 --> 01:14:26.459
Thanks very much, Sandra.
1606
01:14:26.459 --> 01:14:28.040
Wonderful.
1607
01:14:28.040 --> 01:14:33.040
And our next speaker will be Vaughn Steele from Yale.
1608
01:14:33.630 --> 01:14:34.463
Vaughn.
1609
01:14:36.830 --> 01:14:38.027
Hello everybody.
1610
01:14:39.330 --> 01:14:40.820
Can you hear me okay?
1611
01:14:40.820 --> 01:14:41.670
Yes.
1612
01:14:41.670 --> 01:14:42.660
Great.
1613
01:14:42.660 --> 01:14:43.750
It's my pleasure to be here today.
1614
01:14:43.750 --> 01:14:47.570
I wanna thank the committee and the organizers.
1615
01:14:47.570 --> 01:14:49.080
Of course, it's an honor to be here
1616
01:14:49.080 --> 01:14:51.930
to present my program of research.
1617
01:14:51.930 --> 01:14:54.270
So next slide.
1618
01:14:54.270 --> 01:14:56.830
I'm a new investigator,
1619
01:14:56.830 --> 01:14:58.980
so we could switch slides, Dave.
1620
01:14:58.980 --> 01:14:59.930
Thank you.
1621
01:14:59.930 --> 01:15:01.480
There are many people who've helped me get here.
1622
01:15:01.480 --> 01:15:04.160
Most recently, I was a postdoc in Elliot Stein's Lab
1623
01:15:04.160 --> 01:15:05.980
and his group at the NIDA-IRP.
1624
01:15:05.980 --> 01:15:08.400
So they funded much of the work at the backend.
1625
01:15:08.400 --> 01:15:09.340
And I got a grant
1626
01:15:09.340 --> 01:15:11.480
from the Center of Compulsive Behaviors there too.
1627
01:15:11.480 --> 01:15:14.850
I'm Yale psychiatry now on that K-12 and my lab
1628
01:15:14.850 --> 01:15:17.640
is in the Olin Neuropsychiatry Research Center
1629
01:15:17.640 --> 01:15:19.843
that's funded by NIDA and NIAAA.
1630
01:15:20.980 --> 01:15:23.570
I have nothing to disclose, which is my next slide.
1631
01:15:23.570 --> 01:15:26.350
I'm not making any money on anything here.
1632
01:15:26.350 --> 01:15:28.850
So next slide, we'll just get right into it.
1633
01:15:28.850 --> 01:15:31.170
So drug abuse is known to be
1634
01:15:31.170 --> 01:15:34.890
a chronic relapsing brain disease where the field is going.
1635
01:15:34.890 --> 01:15:35.760
This is an exemplar
1636
01:15:35.760 --> 01:15:37.810
from Elliot's Lab NIDA-IRP
1637
01:15:37.810 --> 01:15:41.380
showing functional connectivity dysregulations
1638
01:15:41.380 --> 01:15:42.530
in cocaine users
1639
01:15:42.530 --> 01:15:45.713
inside the mesocorticolimbic dopamine system
1640
01:15:45.713 --> 01:15:49.890
from a resting state functional connectivity analysis.
1641
01:15:49.890 --> 01:15:52.620
This map sum really well the next slide
1642
01:15:52.620 --> 01:15:56.580
of the addiction disease model,
1643
01:15:56.580 --> 01:16:00.730
the Dr. Volkow and Koob spouse and thoroughly explained
1644
01:16:00.730 --> 01:16:05.230
that this cyclical model is agnostic across drugs of abuse.
1645
01:16:05.230 --> 01:16:07.540
It can identify different areas
1646
01:16:07.540 --> 01:16:09.260
in the brain that are implicated
1647
01:16:09.260 --> 01:16:11.400
and different cognitive functions that are dysregulated
1648
01:16:11.400 --> 01:16:13.533
in drugs of abuse.
1649
01:16:14.710 --> 01:16:17.093
Next slide, please.
1650
01:16:20.377 --> 01:16:22.390
It identifies that there are dysregulated circuits
1651
01:16:22.390 --> 01:16:23.223
in the brain.
1652
01:16:23.223 --> 01:16:26.020
And then my idea is that the dysregulation
1653
01:16:26.020 --> 01:16:28.830
could be a target for intervention.
1654
01:16:28.830 --> 01:16:31.190
So if we can identify the cognitive functions
1655
01:16:31.190 --> 01:16:33.610
in the circuits, perhaps we can use
1656
01:16:33.610 --> 01:16:35.300
the neuroplasticity that was used
1657
01:16:35.300 --> 01:16:37.700
to develop substance use disorder,
1658
01:16:37.700 --> 01:16:39.600
we could use that in the back way
1659
01:16:39.600 --> 01:16:42.533
and get them to be use it as a treatment.
1660
01:16:43.860 --> 01:16:46.300
So the big question is is where do we start?
1661
01:16:46.300 --> 01:16:49.127
Which cognitive function in which regions do we target?
1662
01:16:49.127 --> 01:16:50.830
And so I've spent some time
1663
01:16:50.830 --> 01:16:53.500
targeting executive control processes.
1664
01:16:53.500 --> 01:16:57.680
So here is data from a Go/No-Go task that was collected
1665
01:16:58.949 --> 01:16:59.782
before participants initiated
1666
01:16:59.782 --> 01:17:03.540
12-week program substance use treatment.
1667
01:17:03.540 --> 01:17:07.860
Here's this localized or are targeted to the onset
1668
01:17:07.860 --> 01:17:10.247
of the error response.
1669
01:17:10.247 --> 01:17:12.090
There are two processes here,
1670
01:17:12.090 --> 01:17:14.690
the early initial ERN, oh, I made an error,
1671
01:17:14.690 --> 01:17:15.810
and then the secondary process
1672
01:17:15.810 --> 01:17:17.230
of what do I do with this information.
1673
01:17:17.230 --> 01:17:18.860
So a top-down control.
1674
01:17:18.860 --> 01:17:21.500
And you can see what those lines, there are differences
1675
01:17:21.500 --> 01:17:24.050
between individuals who in the red
1676
01:17:24.050 --> 01:17:25.960
actually subsequently completed treatment,
1677
01:17:25.960 --> 01:17:29.370
and then the blues who discontinued the treatment.
1678
01:17:29.370 --> 01:17:31.630
We support vector machine learning models.
1679
01:17:31.630 --> 01:17:34.930
I can predict then with these measures
1680
01:17:34.930 --> 01:17:36.750
who will and won't complete drug treatment.
1681
01:17:36.750 --> 01:17:41.170
So I can predict almost 79% of those individuals
1682
01:17:41.170 --> 01:17:43.480
who will subsequently complete treatment,
1683
01:17:43.480 --> 01:17:46.730
and then 75% who will discontinue treatment.
1684
01:17:46.730 --> 01:17:48.350
And these models also helped me identify
1685
01:17:48.350 --> 01:17:49.493
that it's a secondary process.
1686
01:17:49.493 --> 01:17:52.090
This is executive control top-down control,
1687
01:17:52.090 --> 01:17:55.060
what do I do with this information that is most predictive,
1688
01:17:55.060 --> 01:17:57.010
highlighting that that is a cognitive process
1689
01:17:57.010 --> 01:18:00.489
that we could target for an intervention, to develop
1690
01:18:00.489 --> 01:18:04.083
an intervention that will keep people in treatment longer.
1691
01:18:04.940 --> 01:18:06.210
So what are the neural mechanisms?
1692
01:18:06.210 --> 01:18:09.920
Underneath this is, here's the same Go/No-Go task
1693
01:18:09.920 --> 01:18:13.870
in the scanner.
1694
01:18:13.870 --> 01:18:16.370
And I extracted independent component analysis
1695
01:18:16.370 --> 01:18:17.420
and functional connectivity.
1696
01:18:17.420 --> 01:18:20.560
This is an exemplar of that between anterior singular cortex
1697
01:18:20.560 --> 01:18:21.630
and a sub-cortical IC
1698
01:18:21.630 --> 01:18:24.050
that includes hippocampus and amygdala.
1699
01:18:24.050 --> 01:18:26.370
So when there's dysregulation there
1700
01:18:26.370 --> 01:18:28.120
and individuals who complete versus a discontinue,
1701
01:18:28.120 --> 01:18:30.570
then I can do a pretty good job, even a little bit better
1702
01:18:30.570 --> 01:18:34.460
at predicting an outcome who completes and who discontinues.
1703
01:18:34.460 --> 01:18:36.730
So with this, I've identified a cognitive function
1704
01:18:36.730 --> 01:18:39.650
and circuitry, and what I use in the next slide,
1705
01:18:39.650 --> 01:18:42.160
I just use transcranial magnetic stimulation
1706
01:18:42.160 --> 01:18:45.270
to try to target these cognitive function
1707
01:18:45.270 --> 01:18:48.963
and these circuits that are underlying.
1708
01:18:50.030 --> 01:18:50.993
So next slide.
1709
01:18:51.917 --> 01:18:53.160
And at NIDA, I put together
1710
01:18:53.160 --> 01:18:56.880
a proof-of-concept study using excitatory TMS
1711
01:18:56.880 --> 01:18:58.760
as the intermittent date-of-birth stimulation
1712
01:18:58.760 --> 01:19:01.113
over left dorsolateral prefrontal cortex,
1713
01:19:02.000 --> 01:19:06.110
and had individuals down-regulate their craving
1714
01:19:06.110 --> 01:19:09.387
while viewing cues during the intervention,
1715
01:19:09.387 --> 01:19:10.650
the TMS intervention, with the idea
1716
01:19:10.650 --> 01:19:13.460
that if I can enhance their ability to down-regulate,
1717
01:19:13.460 --> 01:19:17.464
that will, and then the circuitry downstream
1718
01:19:17.464 --> 01:19:19.190
from that could be a potential intervention
1719
01:19:19.190 --> 01:19:23.960
for cocaine use disorder that has no FDA-approved treatment.
1720
01:19:23.960 --> 01:19:26.090
So here I designed a chronic application
1721
01:19:26.090 --> 01:19:29.190
of TMS where we did ITBS three times a day
1722
01:19:29.190 --> 01:19:32.300
for 10 days over two weeks,
1723
01:19:32.300 --> 01:19:34.697
and then a one-week and a four-week follow-up.
1724
01:19:34.697 --> 01:19:36.980
But the main goal is to identify
1725
01:19:36.980 --> 01:19:40.033
and test whether this is actually an intervention
1726
01:19:40.033 --> 01:19:42.673
that could reduce cocaine use.
1727
01:19:43.720 --> 01:19:45.350
So then in the next slide, I can say
1728
01:19:45.350 --> 01:19:47.670
that actually we do have evidence of this
1729
01:19:47.670 --> 01:19:49.370
in the small sample.
1730
01:19:49.370 --> 01:19:51.400
So you can see in both the amount and frequency,
1731
01:19:51.400 --> 01:19:53.040
the bars on the left are baseline,
1732
01:19:53.040 --> 01:19:56.090
and then the next two bars are in each of these figures
1733
01:19:58.167 --> 01:19:59.311
first week and four-week follow-up,
1734
01:19:59.311 --> 01:20:00.530
showing that both amount and frequency
1735
01:20:00.530 --> 01:20:04.023
of cocaine use post-treatment reduced in these individuals,
1736
01:20:05.010 --> 01:20:07.410
suggesting that there's some behavioral change that lasts
1737
01:20:07.410 --> 01:20:08.763
at least a month.
1738
01:20:09.840 --> 01:20:13.080
I can say that every participant reported
1739
01:20:13.080 --> 01:20:14.750
a new interaction with cocaine.
1740
01:20:14.750 --> 01:20:16.560
They all used again, they all reported
1741
01:20:16.560 --> 01:20:18.520
that they couldn't get quite as high
1742
01:20:18.520 --> 01:20:20.033
as they did before.
1743
01:20:22.340 --> 01:20:26.441
They could actually, when they did use,
1744
01:20:26.441 --> 01:20:28.620
they didn't have the same compulsive use,
1745
01:20:28.620 --> 01:20:30.320
they could start and stop.
1746
01:20:30.320 --> 01:20:32.240
They didn't have the same drive to use again.
1747
01:20:32.240 --> 01:20:34.973
So something clearly is changing in these individuals.
1748
01:20:37.690 --> 01:20:39.240
Yeah, all right, so next slide.
1749
01:20:40.840 --> 01:20:42.280
So this is just an exemplar
1750
01:20:42.280 --> 01:20:44.990
of me showing that with machine learning,
1751
01:20:44.990 --> 01:20:47.110
I can identify both cognitive functions
1752
01:20:47.110 --> 01:20:48.890
in neurocircuitry that are dysregulated
1753
01:20:48.890 --> 01:20:51.660
in addiction and develop novel interventions
1754
01:20:51.660 --> 01:20:56.660
and targets for in my case TMS, so what's, that are useful
1755
01:20:56.850 --> 01:20:59.523
in reducing cocaine use behavior.
1756
01:21:01.115 --> 01:21:02.930
This study was a good proof of concept
1757
01:21:02.930 --> 01:21:04.730
and it answered a question of if TMS
1758
01:21:04.730 --> 01:21:08.910
is a viable intervention tool for cocaine use disorder,
1759
01:21:08.910 --> 01:21:11.590
but as any good research study does it,
1760
01:21:11.590 --> 01:21:14.810
it generates new questions, more questions than it answers.
1761
01:21:14.810 --> 01:21:17.650
So here in my program of research at Yale
1762
01:21:17.650 --> 01:21:19.750
as I start my new lab, I am working
1763
01:21:19.750 --> 01:21:23.690
to better understand the neurocircuitry that TMS modulates.
1764
01:21:23.690 --> 01:21:26.114
How long does that modulation last?
1765
01:21:26.114 --> 01:21:31.114
Can it work in other drugs of abuse in addition to cocaine?
1766
01:21:31.270 --> 01:21:35.030
And then also there was no placebo in here.
1767
01:21:35.030 --> 01:21:38.480
What are the placebo effects as there's a large potential
1768
01:21:38.480 --> 01:21:41.690
for a placebo intervention in TMS?
1769
01:21:41.690 --> 01:21:43.340
For that I thank you for your time
1770
01:21:43.340 --> 01:21:45.240
and I'm happy to answer any questions.
1771
01:21:48.270 --> 01:21:50.380
Thank you very much Vaughn.
1772
01:21:50.380 --> 01:21:53.943
Okay, we all have a time for a few questions.
1773
01:21:55.660 --> 01:21:58.140
Did you see any other changes in behavior
1774
01:21:58.140 --> 01:22:01.160
in these patients, especially positive behaviors
1775
01:22:01.160 --> 01:22:03.423
such as improved self-control?
1776
01:22:07.450 --> 01:22:08.283
We did see
1777
01:22:08.283 --> 01:22:11.900
other off, I would call them off-target effects.
1778
01:22:11.900 --> 01:22:13.060
It's a small sample
1779
01:22:13.060 --> 01:22:16.910
so it's hard to make large interpretations,
1780
01:22:16.910 --> 01:22:20.269
but at the other end of it, the individuals that use
1781
01:22:20.269 --> 01:22:23.540
other drugs of abuse, like alcohol, nicotine, marijuana,
1782
01:22:23.540 --> 01:22:25.830
they also reported a reduction in use.
1783
01:22:25.830 --> 01:22:27.317
So there's some of that that's positivity
1784
01:22:27.317 --> 01:22:29.500
and it's a common circuit, so there's some idea
1785
01:22:29.500 --> 01:22:32.350
that that's modulating everything altogether.
1786
01:22:32.350 --> 01:22:35.760
This is the same location that people target
1787
01:22:35.760 --> 01:22:37.600
with TMS to treat depression.
1788
01:22:37.600 --> 01:22:38.921
And though none
1789
01:22:38.921 --> 01:22:40.980
of my participants were clinically depressed,
1790
01:22:40.980 --> 01:22:43.440
they all had a little bit of a pep in their step,
1791
01:22:43.440 --> 01:22:45.870
a little bit happier and mood in day three, day four.
1792
01:22:45.870 --> 01:22:49.730
So there's some other positive side effects
1793
01:22:49.730 --> 01:22:51.820
that are not what I was targeting.
1794
01:22:51.820 --> 01:22:52.653
Yeah.
1795
01:22:54.530 --> 01:22:56.839
And I may have missed it,
1796
01:22:56.839 --> 01:23:00.000
were these long-term users?
1797
01:23:00.000 --> 01:23:01.743
Were they motivated to quit?
1798
01:23:02.680 --> 01:23:06.014
They were non-treatment seeking long-term.
1799
01:23:06.014 --> 01:23:08.370
I think the average is 25-year use.
1800
01:23:08.370 --> 01:23:09.960
At this point Roger,
1801
01:23:09.960 --> 01:23:12.190
I'm sure you know that individuals have gone
1802
01:23:12.190 --> 01:23:14.530
through many, many different treatments and failed
1803
01:23:14.530 --> 01:23:17.840
and they show up and say, you can't fix me,
1804
01:23:17.840 --> 01:23:20.010
then nothing's gonna work, but I'm happy to help
1805
01:23:20.010 --> 01:23:21.310
and maybe you can develop something
1806
01:23:21.310 --> 01:23:24.885
for younger 22-year-olds who are starting.
1807
01:23:24.885 --> 01:23:25.830
These are all people who are like 45-plus,
1808
01:23:25.830 --> 01:23:27.993
they've been doing for 25 years or so.
1809
01:23:29.406 --> 01:23:30.239
But yeah, so it was a little bit
1810
01:23:30.239 --> 01:23:32.360
of a surprise that we had this universal response
1811
01:23:32.360 --> 01:23:34.070
that I'm not here to quit,
1812
01:23:34.070 --> 01:23:35.620
I'm here to help you guys out
1813
01:23:35.620 --> 01:23:38.430
and get paid a little bit all along the way.
1814
01:23:38.430 --> 01:23:40.870
And then they all responded with, it's not the same
1815
01:23:40.870 --> 01:23:43.520
now that I've done this, I don't have the same drive.
1816
01:23:44.730 --> 01:23:45.980
So we really need to understand
1817
01:23:45.980 --> 01:23:48.900
what are the neural mechanisms of change that are happening?
1818
01:23:48.900 --> 01:23:51.970
How do we harness those to develop a treatment?
1819
01:23:51.970 --> 01:23:54.030
And then how long do they last?
1820
01:23:54.030 --> 01:23:55.010
It's an exciting time,
1821
01:23:55.010 --> 01:23:57.543
or new questions that we need to answer.
1822
01:23:58.573 --> 01:23:59.823
Absolutely, absolutely.
1823
01:24:01.870 --> 01:24:02.743
Okay.
1824
01:24:03.580 --> 01:24:04.968
Thanks Vaughn (coughs)
1825
01:24:04.968 --> 01:24:06.147
Thanks to you.
1826
01:24:06.147 --> 01:24:08.390
And John, I'll turn it back to you.
1827
01:24:08.390 --> 01:24:11.680
Yeah, I think I picked up a few questions we missed
1828
01:24:11.680 --> 01:24:13.500
and if anybody has any more questions,
1829
01:24:13.500 --> 01:24:16.070
we have about five minutes.
1830
01:24:16.070 --> 01:24:18.990
It looks like this question we didn't have time
1831
01:24:18.990 --> 01:24:20.593
for is to Kathryn.
1832
01:24:21.890 --> 01:24:22.890
And it's-
Yeah.
1833
01:24:22.890 --> 01:24:23.723
Okay.
1834
01:24:24.807 --> 01:24:25.942
Let's see, let me read the question.
1835
01:24:25.942 --> 01:24:30.220
Even within OUD group, you would need a reward modulation
1836
01:24:30.220 --> 01:24:32.610
specific to drug use.
1837
01:24:32.610 --> 01:24:34.213
How would you deal with that?
1838
01:24:35.814 --> 01:24:38.580
So I'm hoping I'm taking this question correctly.
1839
01:24:38.580 --> 01:24:41.470
So in terms of the different types of drug use,
1840
01:24:41.470 --> 01:24:44.240
do you mean in terms of a treatment?
1841
01:24:44.240 --> 01:24:46.015
So for example, opiod uses
1842
01:24:46.015 --> 01:24:46.950
let's say methadone or buprenorphine,
1843
01:24:46.950 --> 01:24:51.060
in which case we would use the same type of TMS treatment.
1844
01:24:51.060 --> 01:24:52.960
And we could potentially create subgroups
1845
01:24:52.960 --> 01:24:55.510
to say how TMS affects those sub groups.
1846
01:24:55.510 --> 01:24:57.360
Otherwise, another way that we can look
1847
01:24:57.360 --> 01:25:00.080
at drug specific-effects and reward specific effects
1848
01:25:00.080 --> 01:25:02.140
is by providing drug-specific cues
1849
01:25:02.140 --> 01:25:03.880
in different version of the TMS task.
1850
01:25:03.880 --> 01:25:07.610
So for example, providing a monetary reward
1851
01:25:07.610 --> 01:25:08.810
versus heroin reward,
1852
01:25:08.810 --> 01:25:10.980
obviously that would be something quite difficult to do
1853
01:25:10.980 --> 01:25:14.340
as we can't actively give people heroin,
1854
01:25:14.340 --> 01:25:15.560
but we have done something similar
1855
01:25:15.560 --> 01:25:17.810
in smokers where we give them pops of a cigarette.
1856
01:25:17.810 --> 01:25:19.290
So if we could find an analog
1857
01:25:19.290 --> 01:25:21.233
of that kind of reward in money versus drug-specific reward,
1858
01:25:21.233 --> 01:25:25.200
we can start looking at the differential effects
1859
01:25:25.200 --> 01:25:26.890
of hypo or hyperactivity
1860
01:25:26.890 --> 01:25:28.690
to those different types of rewards.
1861
01:25:30.065 --> 01:25:31.530
Okay thank you.
1862
01:25:31.530 --> 01:25:33.600
Another question, and this is actually addressed
1863
01:25:33.600 --> 01:25:35.483
to everyone in the audience.
1864
01:25:36.560 --> 01:25:38.760
Hopefully someone will be able to answer it.
1865
01:25:39.810 --> 01:25:43.743
Does 23andMe disclose addiction?
1866
01:25:50.960 --> 01:25:52.660
I can speak to that for some-
1867
01:25:52.660 --> 01:25:54.014
Sure, please do.
1868
01:25:54.014 --> 01:25:54.910
My camera's not showing,
1869
01:25:54.910 --> 01:25:57.110
but you know how I look.
1870
01:25:57.110 --> 01:25:58.937
You've seen me before.
1871
01:25:58.937 --> 01:25:59.870
(Roger and Sandra laughing)
1872
01:25:59.870 --> 01:26:04.870
So 23andMe has information about self-reported diagnosis.
1873
01:26:07.250 --> 01:26:10.390
So it was not clinical diagnosis, but they have information
1874
01:26:10.390 --> 01:26:14.570
on whether individuals have a substance use disorder.
1875
01:26:14.570 --> 01:26:18.390
There's also some surveys that we have performed
1876
01:26:18.390 --> 01:26:20.290
with Abraham Palmer,
1877
01:26:20.290 --> 01:26:23.040
where we have specifically collected information
1878
01:26:23.040 --> 01:26:25.510
about substance abuse disorders.
1879
01:26:25.510 --> 01:26:27.910
So there is information, but it's self-reported.
1880
01:26:30.340 --> 01:26:31.614
Okay.
1881
01:26:31.614 --> 01:26:32.800
George, do you have a question?
1882
01:26:32.800 --> 01:26:33.760
No, just to comment
1883
01:26:33.760 --> 01:26:38.760
that we also have the AUDIT-C in all of us.
1884
01:26:39.050 --> 01:26:42.310
So we're trying to get the AUDIT-C
1885
01:26:42.310 --> 01:26:46.840
in virtually every cohort at NIH associated
1886
01:26:46.840 --> 01:26:49.270
with the COVID-19 epidemic among other things.
1887
01:26:49.270 --> 01:26:54.270
So I was really, really impressed with the analysis
1888
01:26:55.020 --> 01:26:58.370
of the AUDIT-C talk that.
1889
01:26:58.370 --> 01:27:00.715
That's really exciting-
1890
01:27:00.715 --> 01:27:03.046
That Sandra, that you gave.
1891
01:27:03.046 --> 01:27:08.046
So we will probably be in touch (laughing)
1892
01:27:09.860 --> 01:27:11.560
Great, looking forward.
1893
01:27:13.900 --> 01:27:15.309
Okay, thank you, George.
1894
01:27:15.309 --> 01:27:19.163
Any more questions or comments right now?
1895
01:27:21.530 --> 01:27:24.450
Okay, if not, I certainly like to say that again this year,
1896
01:27:24.450 --> 01:27:27.040
I think the Early Career Investigator Showcase,
1897
01:27:27.040 --> 01:27:29.740
I'm sure my co-chair Roger will agree with me,
1898
01:27:29.740 --> 01:27:31.470
it's just been fantastic.
1899
01:27:31.470 --> 01:27:33.330
I think we should have a lot
1900
01:27:33.330 --> 01:27:36.420
of confidence in the new generation coming up
1901
01:27:36.420 --> 01:27:40.930
and try to support them as much as possible.
1902
01:27:40.930 --> 01:27:44.670
And yeah, I look forward to people next year
1903
01:27:44.670 --> 01:27:47.430
presenting at the showcase in everything.
1904
01:27:47.430 --> 01:27:52.140
I wanna say before we move on to our scientific session,
1905
01:27:52.140 --> 01:27:55.620
which is our last session of the mini-convention,
1906
01:27:55.620 --> 01:27:57.710
right after that last session,
1907
01:27:57.710 --> 01:27:59.280
there will be special comments
1908
01:27:59.280 --> 01:28:01.820
from both Nora Volkow and George Cobb.
1909
01:28:01.820 --> 01:28:04.080
So please stay tuned for that.
1910
01:28:04.080 --> 01:28:06.390
So right now I'd like to move
1911
01:28:06.390 --> 01:28:08.610
to our last session of the day
1912
01:28:09.453 --> 01:28:10.810
which is a scientific session,
1913
01:28:10.810 --> 01:28:14.160
and my colleagues from NIDA, Susan and Vani,
1914
01:28:14.160 --> 01:28:15.210
take it away, please.
1915
01:28:20.210 --> 01:28:21.330
Good afternoon, and welcome
1916
01:28:21.330 --> 01:28:22.760
to our second scientific session
1917
01:28:22.760 --> 01:28:26.180
of this meeting, which explores AI-based approaches
1918
01:28:26.180 --> 01:28:29.270
to addiction, pathophysiology and novel therapeutics.
1919
01:28:29.270 --> 01:28:32.236
My name is Susan Wright and I'll be moderating this session
1920
01:28:32.236 --> 01:28:33.813
with my colleague Vani Pariyadath.
1921
01:28:33.813 --> 01:28:34.646
We were both in the Division
1922
01:28:34.646 --> 01:28:36.790
of Neuroscience and Behavior at NIDA.
1923
01:28:36.790 --> 01:28:37.670
I'm the Program Director
1924
01:28:37.670 --> 01:28:39.320
for Big Data and Computational Science,
1925
01:28:39.320 --> 01:28:41.002
and Vani is the Chief
1926
01:28:41.002 --> 01:28:43.110
of our Behavioral and Cognitive Neuroscience branch.
1927
01:28:43.110 --> 01:28:45.100
We have an excellent lineup of speakers today
1928
01:28:45.100 --> 01:28:47.790
from both academia and industry with topics
1929
01:28:47.790 --> 01:28:50.910
including drug repurposing for opioid use disorders,
1930
01:28:50.910 --> 01:28:53.270
decoding addiction and systemic adaption
1931
01:28:53.270 --> 01:28:56.600
and data-driven approach, predicting various mechanisms
1932
01:28:56.600 --> 01:28:58.470
and using super-computing assistance biology
1933
01:28:58.470 --> 01:29:01.130
to understand complex neural systems.
1934
01:29:01.130 --> 01:29:02.730
The format of the session is that first,
1935
01:29:02.730 --> 01:29:04.410
presentations will take place,
1936
01:29:04.410 --> 01:29:05.840
and then we will have a joint discussion
1937
01:29:05.840 --> 01:29:07.760
and Q&A at the end of the session.
1938
01:29:07.760 --> 01:29:09.720
We ask you to post questions in the chat box,
1939
01:29:09.720 --> 01:29:12.570
and please indicate which speaker your question is for.
1940
01:29:12.570 --> 01:29:15.010
We'll try to get to all the questions during the Q&A,
1941
01:29:15.010 --> 01:29:16.460
but if for some reason we run out of time,
1942
01:29:16.460 --> 01:29:19.023
we'll ask the speakers to respond in the chat box.
1943
01:29:20.260 --> 01:29:22.230
So I'd like to introduce our first speaker,
1944
01:29:22.230 --> 01:29:25.320
Dr. Rong Xu from Case Western Reserve University,
1945
01:29:25.320 --> 01:29:27.710
where she is a Professor of Biomedical Informatics
1946
01:29:27.710 --> 01:29:31.110
and the Founding Director of the newly established
1947
01:29:31.110 --> 01:29:33.420
Center for Artificial Intelligence and Drug Discovery.
1948
01:29:33.420 --> 01:29:34.820
Dr. Xu received her PhD
1949
01:29:34.820 --> 01:29:36.770
in Biomedical Informatics and a Certificate
1950
01:29:36.770 --> 01:29:39.320
in Entrepreneurship from Stanford University.
1951
01:29:39.320 --> 01:29:40.250
Her research focuses
1952
01:29:40.250 --> 01:29:43.640
on revealing the mechanisms that underlie human diseases
1953
01:29:43.640 --> 01:29:45.750
and discover new treatments to combat them.
1954
01:29:45.750 --> 01:29:48.070
And she developed many advanced computational techniques
1955
01:29:48.070 --> 01:29:50.830
to understand, integrate and analyze large amounts
1956
01:29:50.830 --> 01:29:53.310
of complex biological and health data.
1957
01:29:53.310 --> 01:29:54.910
The title of her talk today
1958
01:29:54.910 --> 01:29:58.250
is Drug repurposing for Opioid Use Disorders, Integration
1959
01:29:58.250 --> 01:30:01.260
of Computational Prediction, Clinical Collaboration
1960
01:30:01.260 --> 01:30:04.540
and Mechanism of Action Analysis.
1961
01:30:04.540 --> 01:30:06.130
Take it away Rong.
1962
01:30:06.130 --> 01:30:07.424
Okay.
1963
01:30:07.424 --> 01:30:08.401
Thank you.
1964
01:30:08.401 --> 01:30:09.451
Can you see my slide?
1965
01:30:10.490 --> 01:30:11.590
Yes.
1966
01:30:11.590 --> 01:30:12.423
Oh, okay.
1967
01:30:12.423 --> 01:30:13.623
Thank you.
1968
01:30:13.623 --> 01:30:16.050
So is a great pleasure and honor
1969
01:30:16.050 --> 01:30:21.050
to be here to present our recent work for drug repurposing
1970
01:30:23.130 --> 01:30:27.513
to treat opiod addiction using AI technology.
1971
01:30:30.760 --> 01:30:34.200
So for today's talk, I'm just briefly talk
1972
01:30:34.200 --> 01:30:37.940
about, just briefly describe the overall idea
1973
01:30:37.940 --> 01:30:41.760
of AI-based drug discovery in my group,
1974
01:30:41.760 --> 01:30:46.480
and then also talk about the how we use this kind
1975
01:30:46.480 --> 01:30:51.480
of framework, apply them to a drug purposing
1976
01:30:52.100 --> 01:30:54.083
for opioid addiction.
1977
01:30:58.750 --> 01:31:00.540
So the general research
1978
01:31:00.540 --> 01:31:04.500
in my group is not limited to drug discovery.
1979
01:31:04.500 --> 01:31:09.500
Basically we were interested in broad biomedical discovery,
1980
01:31:09.741 --> 01:31:14.090
including drug discovery, disease understanding,
1981
01:31:14.090 --> 01:31:18.360
also gut microbiome health, gut microbiome related
1982
01:31:18.360 --> 01:31:20.850
to human health and the disease.
1983
01:31:20.850 --> 01:31:22.930
And also how you warm it.
1984
01:31:22.930 --> 01:31:25.700
For example, you warm in the chemicals
1985
01:31:25.700 --> 01:31:28.178
impacted human disease and health.
1986
01:31:28.178 --> 01:31:32.260
Also we were interested in health outcomes
1987
01:31:33.215 --> 01:31:35.740
and cost-effective analysis.
1988
01:31:35.740 --> 01:31:38.390
Well, I hate to interrupt you.
1989
01:31:38.390 --> 01:31:39.223
I don't think your slides are advancing.
1990
01:31:41.742 --> 01:31:42.743
Oh
1991
01:31:42.743 --> 01:31:43.703
So let me say in first slide.
1992
01:31:44.590 --> 01:31:45.920
We're actually seeing a PDF
1993
01:31:45.920 --> 01:31:48.220
of your slides, not PowerPoint.
1994
01:31:48.220 --> 01:31:49.463
Yeah, is a PDF.
1995
01:31:54.070 --> 01:31:55.740
Is advancing?
1996
01:31:55.740 --> 01:31:59.560
Is called the research interest in our group, the title.
1997
01:31:59.560 --> 01:32:02.560
It's still showing that title page.
1998
01:32:02.560 --> 01:32:03.623
Oh, what?
1999
01:32:04.550 --> 01:32:05.810
Sorry about that.
2000
01:32:05.810 --> 01:32:06.860
The sharing is...
2001
01:32:16.305 --> 01:32:18.138
Let me share it, okay.
2002
01:32:33.430 --> 01:32:34.453
Sorry about that.
2003
01:32:35.360 --> 01:32:37.330
Doctor, I have your slides here.
2004
01:32:37.330 --> 01:32:38.785
This is Dave.
2005
01:32:38.785 --> 01:32:41.092
I can share them if you want me to.
2006
01:32:41.092 --> 01:32:43.530
I updated a little bit.
2007
01:32:43.530 --> 01:32:44.583
Is now showing?
2008
01:32:45.660 --> 01:32:47.890
The second page called you (indistinct)
2009
01:32:47.890 --> 01:32:50.440
Okay, you stopped sharing your screen, so.
2010
01:33:08.950 --> 01:33:09.783
Is sharing?
2011
01:33:12.390 --> 01:33:13.680
Yes.
2012
01:33:13.680 --> 01:33:15.903
Okay, so this, the second page.
2013
01:33:28.390 --> 01:33:33.390
Yeah, so it said they still paused
2014
01:33:35.460 --> 01:33:36.930
It's on the second slide now
2015
01:33:36.930 --> 01:33:40.066
if you would like to continue.
Oh, okay.
2016
01:33:40.066 --> 01:33:42.480
So I just talk
2017
01:33:42.480 --> 01:33:45.520
about the overall research in my group.
2018
01:33:45.520 --> 01:33:48.490
Basically we use the web for all kinds
2019
01:33:48.490 --> 01:33:51.480
of computational algorithms,
2020
01:33:51.480 --> 01:33:54.090
including intelligence technology
2021
01:33:54.090 --> 01:33:57.410
for broad biomedical discovery
2022
01:33:57.410 --> 01:33:59.363
including drug discovery.
2023
01:34:00.420 --> 01:34:02.860
So today I'm just briefly talk
2024
01:34:02.860 --> 01:34:07.559
about the AI-based the drug discovery, the general idea
2025
01:34:07.559 --> 01:34:09.840
for our research.
2026
01:34:09.840 --> 01:34:12.130
So basically the first idea here
2027
01:34:12.130 --> 01:34:16.770
is we want to for drug discovery, different
2028
01:34:16.770 --> 01:34:18.870
from traditional drug discovery,
2029
01:34:18.870 --> 01:34:22.453
which is usually, you identify,
2030
01:34:23.295 --> 01:34:26.820
like you need the compounds in experiment models,
2031
01:34:26.820 --> 01:34:29.060
then you do clinical trials.
2032
01:34:29.060 --> 01:34:32.570
The problem with studies animal model is working
2033
01:34:32.570 --> 01:34:34.709
from the human in humans.
2034
01:34:34.709 --> 01:34:37.959
So this is why like over 90% drugs
2035
01:34:39.175 --> 01:34:42.083
work with the animal model but they failed
2036
01:34:42.083 --> 01:34:43.667
in clinical trials.
2037
01:34:43.667 --> 01:34:46.530
So what we're trying to do is we want to incorporate
2038
01:34:46.530 --> 01:34:50.645
much amount of human phenotype data in the very early stage
2039
01:34:50.645 --> 01:34:53.763
of drug discovery.
2040
01:34:54.925 --> 01:34:59.925
So how we're going to do that?
2041
01:35:06.301 --> 01:35:07.852
So how we're going to do that.
2042
01:35:07.852 --> 01:35:12.690
So we basically with the AI technology
2043
01:35:12.690 --> 01:35:15.660
to extract a phenotypical data
2044
01:35:15.660 --> 01:35:20.660
from all published biomedical research articles
2045
01:35:21.240 --> 01:35:24.200
from patient electronic medical record,
2046
01:35:24.200 --> 01:35:27.790
and develop a technology to understand the integrator
2047
01:35:27.790 --> 01:35:32.310
and the analyze this kind of complex phenotypical data
2048
01:35:32.310 --> 01:35:33.532
and then integrate
2049
01:35:33.532 --> 01:35:38.300
with public- available genetics and genomics data.
2050
01:35:38.300 --> 01:35:39.493
So-
2051
01:35:39.493 --> 01:35:40.730
Rong, these slides are not advancing,
2052
01:35:40.730 --> 01:35:42.560
so we're gonna have Dave take
2053
01:35:42.560 --> 01:35:45.100
over the sides for the version he has.
2054
01:35:45.100 --> 01:35:49.160
Oh, can I send you the (indistinct) notes (laughing)
2055
01:35:50.550 --> 01:35:52.362
I think for the sake of time
2056
01:35:52.362 --> 01:35:53.670
we're just gonna have to go with what he already has.
2057
01:35:53.670 --> 01:35:54.703
Okay, yeah.
2058
01:35:55.700 --> 01:35:56.533
Great.
2059
01:35:58.006 --> 01:35:58.839
Right.
2060
01:36:06.320 --> 01:36:08.250
Okay, do you see them now?
2061
01:36:08.250 --> 01:36:10.360
Oh yes, sorry about that.
2062
01:36:10.360 --> 01:36:12.028
No, no, no, you're fine.
2063
01:36:12.028 --> 01:36:13.033
Don't worry.
2064
01:36:13.033 --> 01:36:14.128
Should I be on these-
2065
01:36:14.128 --> 01:36:15.610
Oh yeah, I think with some slides, yeah.
2066
01:36:15.610 --> 01:36:20.300
So then basically so the idea
2067
01:36:20.300 --> 01:36:24.600
is for drug repurposing because it's all FDA-approved drug.
2068
01:36:24.600 --> 01:36:26.180
A lot of people use it,
2069
01:36:26.180 --> 01:36:30.180
even though it's from different disease indications,
2070
01:36:30.180 --> 01:36:31.490
but then we can analyze
2071
01:36:31.490 --> 01:36:35.536
the patient electronic medical record to see,
2072
01:36:35.536 --> 01:36:39.460
to evaluate if the drug have efficacy
2073
01:36:39.460 --> 01:36:41.623
and not a disease indication.
2074
01:36:42.840 --> 01:36:46.020
For example, like opioid addiction
2075
01:36:46.020 --> 01:36:48.153
and other, like depression.
2076
01:36:50.050 --> 01:36:52.393
So next slide, please.
2077
01:36:53.440 --> 01:36:58.440
And then we talked about all the computational reasoning
2078
01:36:58.560 --> 01:37:02.346
by the way also is very important that during this kind
2079
01:37:02.346 --> 01:37:06.930
of a process, we want to integrate the inputs
2080
01:37:06.930 --> 01:37:10.710
from like a human-like experts
2081
01:37:11.910 --> 01:37:16.910
and also interacting with some experimental scientists
2082
01:37:17.460 --> 01:37:21.090
to try and to incorporate the results
2083
01:37:21.090 --> 01:37:22.990
from experiments, the models
2084
01:37:24.338 --> 01:37:26.890
and iterate truly to improve the prediction
2085
01:37:26.890 --> 01:37:28.483
for candidate drugs.
2086
01:37:29.780 --> 01:37:30.863
Next slide.
2087
01:37:32.550 --> 01:37:37.550
So basically, so the most important, and also probably
2088
01:37:38.760 --> 01:37:42.340
the most difficult part of the whole process
2089
01:37:42.340 --> 01:37:44.560
is you want build the brain power
2090
01:37:44.560 --> 01:37:47.100
of the reasoning machine.
2091
01:37:47.100 --> 01:37:49.040
So basically you will want to build
2092
01:37:49.040 --> 01:37:51.880
like a virtual, like AI-reasoning machine,
2093
01:37:51.880 --> 01:37:53.980
which you can read all kinds
2094
01:37:53.980 --> 01:37:58.650
of articles, integrate different evidence together
2095
01:37:58.650 --> 01:38:02.790
and come up with like some (indistinct) hypothesis
2096
01:38:04.650 --> 01:38:06.573
Next slide, please.
2097
01:38:08.411 --> 01:38:11.583
And then, so then the knowledge
2098
01:38:12.989 --> 01:38:16.875
in the biomedical domain also called the wisdom
2099
01:38:16.875 --> 01:38:20.620
of the cloud probably most of them you in there,
2100
01:38:20.620 --> 01:38:24.014
were published by medical research articles.
2101
01:38:24.014 --> 01:38:28.756
Right now we already in the public in the (indistinct)
2102
01:38:28.756 --> 01:38:30.290
we have over 30 million more
2103
01:38:30.290 --> 01:38:32.710
for biomedical research articles.
2104
01:38:32.710 --> 01:38:35.470
If you search like a substance use disorder,
2105
01:38:35.470 --> 01:38:38.050
you'll get a hundreds of thousands of articles.
2106
01:38:38.050 --> 01:38:43.050
And also people study disease sometimes in context
2107
01:38:44.490 --> 01:38:45.500
or not other diseases.
2108
01:38:45.500 --> 01:38:49.170
So from these articles, you can get a lot
2109
01:38:49.170 --> 01:38:50.580
of like a relationship
2110
01:38:50.580 --> 01:38:53.060
among different biomedical entities.
2111
01:38:53.060 --> 01:38:56.530
And also we will also have a lot
2112
01:38:56.530 --> 01:38:59.293
of like patient electronic health records.
2113
01:39:00.690 --> 01:39:02.790
So with all the process and the computer
2114
01:39:02.790 --> 01:39:05.640
cannot even understand the single word
2115
01:39:05.640 --> 01:39:09.590
or what we write or what we a say.
2116
01:39:09.590 --> 01:39:12.983
So the very first step, important that has to use,
2117
01:39:12.983 --> 01:39:17.983
we want to extract, make the knowledge embedded
2118
01:39:18.120 --> 01:39:22.320
in these free text articles too for the make it
2119
01:39:22.320 --> 01:39:26.470
machine-understandable and actionable knowledge.
2120
01:39:26.470 --> 01:39:29.083
So then, next slide, please.
2121
01:39:31.070 --> 01:39:33.480
So the main technology is called
2122
01:39:33.480 --> 01:39:36.120
the natural language processing technology.
2123
01:39:36.120 --> 01:39:40.700
Basically this technology is to enable computers
2124
01:39:40.700 --> 01:39:44.350
to derive meaning from human input,
2125
01:39:44.350 --> 01:39:47.330
and also, so natural language processing
2126
01:39:47.330 --> 01:39:52.330
probably is one of the hardest AI problems.
2127
01:39:52.450 --> 01:39:57.090
Basically we want make a computers as intelligent
2128
01:39:57.090 --> 01:40:00.010
as people as I described before,
2129
01:40:00.010 --> 01:40:04.340
because we want the computer can just read all the articles,
2130
01:40:04.340 --> 01:40:05.550
synthesize knowledge and the come up
2131
01:40:05.550 --> 01:40:09.448
with some Lovell hypothesis, suggests like research,
2132
01:40:09.448 --> 01:40:14.448
read the articles to come up with testable hypothesis.
2133
01:40:14.930 --> 01:40:18.790
And probably different
2134
01:40:18.790 --> 01:40:20.170
from right now (indistinct)
2135
01:40:20.170 --> 01:40:24.050
on the machine learning approach,
2136
01:40:24.050 --> 01:40:28.170
natural language processing usually we cannot solve
2137
01:40:28.170 --> 01:40:30.500
by one single special algorithm.
2138
01:40:30.500 --> 01:40:33.200
Basically different tasks to different data,
2139
01:40:33.200 --> 01:40:35.460
and also based on your requirement
2140
01:40:36.563 --> 01:40:38.250
about the precision where recall Palmer,
2141
01:40:38.250 --> 01:40:42.023
some often you want to come up with different algorithms.
2142
01:40:43.233 --> 01:40:44.193
So next slide.
2143
01:40:46.320 --> 01:40:50.930
So we already have natural language processing techniques
2144
01:40:50.930 --> 01:40:55.930
and other AI technology to, and to construct
2145
01:40:56.076 --> 01:41:00.100
a disease phenom knowledge base,
2146
01:41:00.100 --> 01:41:03.050
from (indistinct) articles,
2147
01:41:03.050 --> 01:41:05.810
their health records over 30 million patients
2148
01:41:05.810 --> 01:41:08.710
and also clinical trial reports.
2149
01:41:08.710 --> 01:41:11.320
Basically in this knowledge base will have
2150
01:41:11.320 --> 01:41:14.700
like this comorbidity relationship
2151
01:41:14.700 --> 01:41:17.510
with disease, disease causal relationship
2152
01:41:17.510 --> 01:41:20.620
and also disease manifestation.
2153
01:41:20.620 --> 01:41:23.666
So this knowledge mostly extracted
2154
01:41:23.666 --> 01:41:25.890
from like free-texts articles.
2155
01:41:25.890 --> 01:41:27.373
So next slide please.
2156
01:41:29.730 --> 01:41:32.932
So this is one of the sub-component
2157
01:41:32.932 --> 01:41:33.765
of this knowledge base
2158
01:41:33.765 --> 01:41:36.470
called disease risk knowledge base.
2159
01:41:36.470 --> 01:41:40.050
So in this knowledge base (indistinct)
2160
01:41:43.220 --> 01:41:44.803
disease causal pairs.
2161
01:41:46.250 --> 01:41:48.670
34,000 diseases causal pair
2162
01:41:48.670 --> 01:41:52.310
for 12,000 unique disease.
2163
01:41:52.310 --> 01:41:55.740
So this, and is highly precise,
2164
01:41:55.740 --> 01:42:00.010
basically will analyze the patterns of how research
2165
01:42:00.010 --> 01:42:02.510
when the right articles or how they describe
2166
01:42:02.510 --> 01:42:05.313
the causal relationship in the writing.
2167
01:42:06.406 --> 01:42:10.100
So this is snapshot, it means what kind of disease
2168
01:42:10.100 --> 01:42:11.880
can cause dementia?
2169
01:42:11.880 --> 01:42:14.338
And also is is automatically weighted
2170
01:42:14.338 --> 01:42:18.963
based on the evidence from the research articles.
2171
01:42:20.320 --> 01:42:22.343
So next slide please.
2172
01:42:24.120 --> 01:42:25.670
And the will also construct
2173
01:42:25.670 --> 01:42:30.670
like a disease phenotype relationship
2174
01:42:31.580 --> 01:42:34.007
which is like a disease side effect,
2175
01:42:35.100 --> 01:42:37.296
that disease treatment
2176
01:42:37.296 --> 01:42:41.260
also from (indistinct) research articles
2177
01:42:41.260 --> 01:42:45.270
and the patient data and the FDA drug labels.
2178
01:42:45.270 --> 01:42:46.483
Next slide please.
2179
01:42:47.920 --> 01:42:50.250
So basically this is snapshot would say,
2180
01:42:50.250 --> 01:42:53.550
you want to say, I have a medical condition,
2181
01:42:53.550 --> 01:42:57.130
for example, depression, the suicide or dementia,
2182
01:42:57.130 --> 01:42:59.897
you wanna say, what kind
2183
01:42:59.897 --> 01:43:02.810
of drugs can potentially cause this kind
2184
01:43:02.810 --> 01:43:04.150
of medical condition.
2185
01:43:04.150 --> 01:43:05.513
Next slide please.
2186
01:43:07.740 --> 01:43:09.580
So basically we extract this kind
2187
01:43:09.580 --> 01:43:14.580
of phenotypical relationship from research articles.
2188
01:43:14.730 --> 01:43:18.630
And then the next important part is to how to integrate
2189
01:43:18.630 --> 01:43:21.050
with publicly available only data
2190
01:43:21.050 --> 01:43:23.320
and the patient electronic records.
2191
01:43:23.320 --> 01:43:24.663
Next slide please.
2192
01:43:26.680 --> 01:43:31.680
So we develop like a context-sensitive network approach
2193
01:43:32.270 --> 01:43:35.117
to try to modeling the heterogeneous
2194
01:43:35.117 --> 01:43:40.117
and the context-specific relationship amongst of hundreds
2195
01:43:40.180 --> 01:43:42.680
of thousand biomedical entities, for example,
2196
01:43:42.680 --> 01:43:45.820
the genes pathways, disease drugs.
2197
01:43:45.820 --> 01:43:46.983
Next slide please.
2198
01:43:49.200 --> 01:43:53.970
So we use this kind of framework for opiod addiction drug
2199
01:43:56.680 --> 01:43:59.080
or repurposing for opiod addiction.
2200
01:43:59.080 --> 01:44:01.211
Basically, next.
2201
01:44:01.211 --> 01:44:04.443
Next slide, please.
2202
01:44:06.390 --> 01:44:11.390
So the first step was so we used drug phenotype data,
2203
01:44:12.040 --> 01:44:15.640
and used the context-sensitive
2204
01:44:16.990 --> 01:44:21.900
network- based approach to first to predict a candidate drug
2205
01:44:21.900 --> 01:44:24.800
for opioid addiction.
2206
01:44:24.800 --> 01:44:26.960
The next we, for top
2207
01:44:28.211 --> 01:44:31.630
identify the candidates.
2208
01:44:31.630 --> 01:44:34.308
We used patient electronic records to analyze
2209
01:44:34.308 --> 01:44:36.610
if this candidate
2210
01:44:36.610 --> 01:44:41.560
really can reduce opioid dependency in patients.
2211
01:44:41.560 --> 01:44:46.180
And then after that, we tried to understand
2212
01:44:46.180 --> 01:44:48.480
the mechanism of action of the candidate
2213
01:44:48.480 --> 01:44:53.167
or why they have potential therapeutic effects
2214
01:44:54.040 --> 01:44:55.950
for opiod dependence.
2215
01:44:55.950 --> 01:44:57.213
Next slide please.
2216
01:44:59.040 --> 01:45:00.140
Yeah, so this one,
2217
01:45:00.140 --> 01:45:04.060
so the first step
2218
01:45:04.060 --> 01:45:06.520
is we use drug phenotype
2219
01:45:06.520 --> 01:45:08.105
and also we also use
2220
01:45:08.105 --> 01:45:13.105
the drug drinking interaction
2221
01:45:15.990 --> 01:45:18.527
on the drug targeted information to try
2222
01:45:18.527 --> 01:45:23.527
and to come up drugs to treat opioid addiction.
2223
01:45:26.200 --> 01:45:28.033
Next slide please.
2224
01:45:29.640 --> 01:45:32.130
And then for, we have access
2225
01:45:32.130 --> 01:45:35.430
to a patient electronic health record
2226
01:45:35.430 --> 01:45:37.740
of 73 million patients.
2227
01:45:37.740 --> 01:45:41.923
We analyzed the top 20 predicted candidate drugs.
2228
01:45:43.260 --> 01:45:45.820
So exposure, so each of candidate drug,
2229
01:45:45.820 --> 01:45:50.430
so the patients is, people already have opioid dependence
2230
01:45:51.882 --> 01:45:53.351
and also have their original indication
2231
01:45:53.351 --> 01:45:54.250
for the candidate drug.
2232
01:45:54.250 --> 01:45:57.664
And then the outcome is we want to see if the drug
2233
01:45:57.664 --> 01:46:00.290
have any efficacy,
2234
01:46:00.290 --> 01:46:03.950
has the remission opioid dependence remission
2235
01:46:03.950 --> 01:46:07.450
in their patients.
2236
01:46:07.450 --> 01:46:09.273
So next slide please.
2237
01:46:10.550 --> 01:46:14.960
And then we obtained a list of opiod addiction genes
2238
01:46:16.288 --> 01:46:18.550
and they have founded the pathways
2239
01:46:18.550 --> 01:46:21.820
and then to see if each of the drugs,
2240
01:46:21.820 --> 01:46:26.820
if the target opioid specific genes and the pathways.
2241
01:46:27.420 --> 01:46:28.440
Next slide, please
2242
01:46:30.140 --> 01:46:33.508
As this result, so first for evaluation
2243
01:46:33.508 --> 01:46:37.120
where you evaluate the algorithm, we found that this drug,
2244
01:46:37.120 --> 01:46:40.630
this algorithm can rank what we know,
2245
01:46:40.630 --> 01:46:45.630
the FDA-approved opioid addiction medication highly
2246
01:46:46.130 --> 01:46:48.970
among all the FDA-approved drugs.
2247
01:46:48.970 --> 01:46:50.313
Next slide, please.
2248
01:46:52.490 --> 01:46:55.380
And then we test the top 20 drugs.
2249
01:46:55.380 --> 01:47:00.280
So on the right-hand side, there is drugs
2250
01:47:00.280 --> 01:47:05.280
could enhance the opioid dependence remission,
2251
01:47:05.570 --> 01:47:08.490
and the (indistinct) makes it worse.
2252
01:47:08.490 --> 01:47:12.580
So this one is what we know is called (indistinct)
2253
01:47:12.580 --> 01:47:14.797
which is highly, so can improve
2254
01:47:16.510 --> 01:47:21.510
the remission means it's come evaluate the algorithm.
2255
01:47:23.330 --> 01:47:26.200
And then there are some other unknown drugs.
2256
01:47:26.200 --> 01:47:29.565
The level-candidate drugs also have potential
2257
01:47:29.565 --> 01:47:33.250
to increase the, enhanced the opioid dependence remission.
2258
01:47:33.250 --> 01:47:34.633
Next slide, please.
2259
01:47:35.980 --> 01:47:38.273
And then we analyzed the top five drug,
2260
01:47:38.273 --> 01:47:39.940
then we found that these drugs
2261
01:47:39.940 --> 01:47:42.493
had many opioid-specificity genes.
2262
01:47:43.580 --> 01:47:44.943
Next slide, please.
2263
01:47:46.240 --> 01:47:48.400
And also we found that this drug targeted
2264
01:47:48.400 --> 01:47:52.170
many opioid-specific genetical pathways.
2265
01:47:52.170 --> 01:47:53.403
Next slide, please.
2266
01:47:55.090 --> 01:47:55.923
Next.
2267
01:47:57.340 --> 01:48:00.990
So the overall summary is we want,
2268
01:48:00.990 --> 01:48:04.350
the longterm goal is we want to develop technologies
2269
01:48:05.220 --> 01:48:09.880
which is like knowledge-driven and also one
2270
01:48:09.880 --> 01:48:14.880
that is explainable which we interact with human,
2271
01:48:15.280 --> 01:48:19.160
make it experts and they experiment models.
2272
01:48:19.160 --> 01:48:24.160
One to, you (indistinct) to build
2273
01:48:24.330 --> 01:48:29.330
their special specialized drug discovery platform
2274
01:48:30.050 --> 01:48:32.410
for specialty diseases.
2275
01:48:32.410 --> 01:48:33.683
Next slide, please.
2276
01:48:34.920 --> 01:48:39.920
So we used this platform is we still in the infant stage.
2277
01:48:41.150 --> 01:48:43.960
We used this platform for opiod addiction
2278
01:48:43.960 --> 01:48:46.930
where I identified so candidate drugs.
2279
01:48:46.930 --> 01:48:51.523
Next step we want to refine top candidate drugs.
2280
01:48:51.523 --> 01:48:54.310
We want to improve the prediction way,
2281
01:48:54.310 --> 01:48:59.090
right now is very simple computation predictions.
2282
01:48:59.090 --> 01:49:04.090
And also want to collaborate with experiment scientist
2283
01:49:04.350 --> 01:49:08.560
to test those drug candidates because is FDA-approved,
2284
01:49:08.560 --> 01:49:11.610
maybe we can launch small clinical trials
2285
01:49:11.610 --> 01:49:16.357
for those candidates, and also this platform
2286
01:49:18.017 --> 01:49:22.210
can in general can apply to other diseases,
2287
01:49:22.210 --> 01:49:25.270
including other substance use disorders.
2288
01:49:25.270 --> 01:49:26.623
Next slide please.
2289
01:49:28.380 --> 01:49:32.420
So these people in my group, I wanna thank them,
2290
01:49:32.420 --> 01:49:37.420
my collaborators, and the yeah, next slide, please.
2291
01:49:39.560 --> 01:49:43.880
My funding support from NIH and from other foundations
2292
01:49:43.880 --> 01:49:46.700
and from the pharmaceutical companies.
2293
01:49:46.700 --> 01:49:47.933
Next slide please.
2294
01:49:49.253 --> 01:49:53.270
Especially from the night the NIDA Clinical Trial Network
2295
01:49:54.342 --> 01:49:57.495
and several project, and the co-investigator
2296
01:49:57.495 --> 01:49:58.750
and the consultant.
2297
01:49:58.750 --> 01:50:00.211
Next slide.
2298
01:50:00.211 --> 01:50:01.215
Oh yeah, that's all.
2299
01:50:01.215 --> 01:50:02.048
Thank you.
2300
01:50:02.048 --> 01:50:05.053
Sorry about this, yeah.
2301
01:50:06.491 --> 01:50:08.100
Thank you Dr. Xu.
2302
01:50:08.100 --> 01:50:11.420
Our next speaker today is Dr. Niven Narain,
2303
01:50:11.420 --> 01:50:16.420
who is the Co-founder and President and CEO of Berg LLC,
2304
01:50:16.770 --> 01:50:18.480
just based in Boston.
2305
01:50:18.480 --> 01:50:20.850
His work has focused on combining biology
2306
01:50:20.850 --> 01:50:24.660
and Bayesian AI towards next-generation drug development.
2307
01:50:24.660 --> 01:50:27.510
And today he'll be talking about engaging deep biology
2308
01:50:27.510 --> 01:50:29.940
and Bayesian AI to decode addiction
2309
01:50:29.940 --> 01:50:33.080
and systemic adaption in a data-driven approach.
2310
01:50:33.080 --> 01:50:35.520
With that, I'll pass this on to Dr. Narain.
2311
01:50:41.673 --> 01:50:42.550
Thank you.
2312
01:50:42.550 --> 01:50:46.030
I'd like to thank NIDA for the invitation
2313
01:50:46.030 --> 01:50:46.863
to give this lecture.
2314
01:50:46.863 --> 01:50:48.200
I think it's really an important time
2315
01:50:48.200 --> 01:50:51.971
because I remember when I received the initial invitation
2316
01:50:51.971 --> 01:50:56.420
about this is lecture and the session, we had just got
2317
01:50:56.420 --> 01:51:00.193
into the early onset of COVID, and since then,
2318
01:51:01.070 --> 01:51:03.810
we've actually experienced unfortunately
2319
01:51:03.810 --> 01:51:06.480
a deeper sensitization and an amplification
2320
01:51:06.480 --> 01:51:08.090
of what addiction has done
2321
01:51:08.090 --> 01:51:13.090
under certain compressed stressed environments like COVID.
2322
01:51:13.800 --> 01:51:18.800
So this is really for Berg,
2323
01:51:19.280 --> 01:51:22.167
in addition to some of our other work
2324
01:51:22.167 --> 01:51:24.730
in brain-related diseases
2325
01:51:24.730 --> 01:51:26.570
such as Parkinson's and Alzheimer's,
2326
01:51:26.570 --> 01:51:30.010
and looking at the metabolomics and lipidomics associated
2327
01:51:30.010 --> 01:51:33.240
with aging to really take an approach
2328
01:51:33.240 --> 01:51:37.440
where we can seek to sensitize the full narrative
2329
01:51:37.440 --> 01:51:42.440
of the biology and engendering the full demographic data,
2330
01:51:42.570 --> 01:51:44.360
real-world evidence data,
2331
01:51:44.360 --> 01:51:47.090
clinical but importantly, all-mix data
2332
01:51:47.090 --> 01:51:51.690
across a long tattooed of either a temporal sampling
2333
01:51:51.690 --> 01:51:55.430
or different types of tissue sets.
2334
01:51:55.430 --> 01:52:00.363
So with that, I will just move directly into my talk.
2335
01:52:02.950 --> 01:52:05.070
So over the past year,
2336
01:52:05.070 --> 01:52:08.610
we've all I think in this session,
2337
01:52:08.610 --> 01:52:10.060
fundamentally appreciate this,
2338
01:52:10.060 --> 01:52:13.100
but especially when it pertains to addiction
2339
01:52:13.100 --> 01:52:16.010
and some of the multifactorial components that are outside
2340
01:52:16.010 --> 01:52:17.400
of the biology and outside
2341
01:52:17.400 --> 01:52:19.600
of the clinical parameterization.
2342
01:52:19.600 --> 01:52:23.840
When you look at demographic and social care components,
2343
01:52:25.170 --> 01:52:27.470
the diversity goes even further,
2344
01:52:27.470 --> 01:52:29.090
the complexity goes even further.
2345
01:52:29.090 --> 01:52:31.640
So I think that the humility really dictates
2346
01:52:31.640 --> 01:52:35.670
how the curation of the data sets are arrived at,
2347
01:52:35.670 --> 01:52:40.040
collaboration, and specially in this type of environment,
2348
01:52:40.040 --> 01:52:42.240
it's so important.
2349
01:52:42.240 --> 01:52:45.910
And I also just wanna just take one moment just to step back
2350
01:52:45.910 --> 01:52:47.790
because I think that AI
2351
01:52:47.790 --> 01:52:52.287
over the past few years has been hyped
2352
01:52:52.287 --> 01:52:55.310
to some extent.
2353
01:52:55.310 --> 01:52:58.041
It is just a level set.
2354
01:52:58.041 --> 01:52:59.830
What we're discussing here
2355
01:52:59.830 --> 01:53:02.320
is a serious problem of addiction.
2356
01:53:02.320 --> 01:53:03.896
We're discussing this,
2357
01:53:03.896 --> 01:53:07.793
maybe one of our most precious organs.
2358
01:53:08.630 --> 01:53:09.760
AI is a tool.
2359
01:53:09.760 --> 01:53:14.760
It's a tool to bring together disparate data sets to help us
2360
01:53:15.690 --> 01:53:19.210
make sense of differential phenotypes to help us make sense
2361
01:53:19.210 --> 01:53:23.390
of controlled data sets, et cetera.
2362
01:53:23.390 --> 01:53:24.640
And not withstanding the fact
2363
01:53:24.640 --> 01:53:26.710
that there are different types of AI,
2364
01:53:26.710 --> 01:53:28.700
whether you're looking at machine learning based
2365
01:53:28.700 --> 01:53:33.300
on priors or neural networks that are gonna look
2366
01:53:33.300 --> 01:53:38.300
at existing data, or Bayesian AI in the case
2367
01:53:39.770 --> 01:53:43.460
of Berg, that's what we've been focused on
2368
01:53:43.460 --> 01:53:47.260
because we wanna allow the differential concentricity
2369
01:53:47.260 --> 01:53:49.330
of that data to take hold
2370
01:53:49.330 --> 01:53:54.163
and just to have a very rudimentary example
2371
01:53:54.163 --> 01:53:58.113
on Bayesian AI with respect to the life sciences.
2372
01:53:59.060 --> 01:54:02.440
I think we all know the very simple component
2373
01:54:02.440 --> 01:54:07.440
of, sorry, of a sprinkler in the rain.
2374
01:54:08.400 --> 01:54:11.455
When it rains, the grass is obviously gonna get red,
2375
01:54:11.455 --> 01:54:16.455
but it's, we learn, we adapt ourselves to the reality
2376
01:54:16.480 --> 01:54:19.371
that when a sprinkler is on, it's gonna inhibit
2377
01:54:19.371 --> 01:54:24.371
the need for for certain components
2378
01:54:25.230 --> 01:54:27.610
along the rain-and-sprinkler axes.
2379
01:54:27.610 --> 01:54:31.570
We're able to tell that the difference over time.
2380
01:54:31.570 --> 01:54:35.210
But if we superimpose that a very simple construct
2381
01:54:35.210 --> 01:54:40.210
upon biological pathways, it has an absence of these types
2382
01:54:41.990 --> 01:54:46.070
of massive data sets and super-compute capability
2383
01:54:46.070 --> 01:54:49.943
and Bayesian-type algorithms, we've been over time,
2384
01:54:52.520 --> 01:54:55.700
it's taken us oftentimes years
2385
01:54:55.700 --> 01:54:57.700
and decades to unravel
2386
01:54:57.700 --> 01:55:00.480
and decipher some hallmark mechanisms.
2387
01:55:00.480 --> 01:55:03.950
And obviously with the advent of these technologies,
2388
01:55:03.950 --> 01:55:08.080
we're seeing the ability
2389
01:55:08.080 --> 01:55:12.210
for us in the scientific community to do this much faster.
2390
01:55:12.210 --> 01:55:17.157
So the approach that Berg has taken is really to embrace
2391
01:55:18.790 --> 01:55:21.190
and encumber different types of data sets
2392
01:55:21.190 --> 01:55:25.290
that do come from, especially in this case,
2393
01:55:25.290 --> 01:55:26.880
environment and outcomes.
2394
01:55:26.880 --> 01:55:29.510
So not only the epigenetic components,
2395
01:55:29.510 --> 01:55:33.660
but as I spoke to, clinical and demographic data.
2396
01:55:33.660 --> 01:55:34.960
And what we're really interested in
2397
01:55:34.960 --> 01:55:39.660
is how in an acute phase measurable outcome,
2398
01:55:39.660 --> 01:55:44.660
not necessarily in urgent care or ER setting,
2399
01:55:44.760 --> 01:55:46.530
we're not gonna be able to screen
2400
01:55:46.530 --> 01:55:51.046
for a genetic mutation, but it is conceivable
2401
01:55:51.046 --> 01:55:55.920
that in some high-throughput centers around the country,
2402
01:55:55.920 --> 01:56:00.750
that we could be a few years away from looking
2403
01:56:00.750 --> 01:56:05.750
at metabolomics and lipidomics and proteomics very quickly,
2404
01:56:06.350 --> 01:56:09.760
but having the basis of signatures of addiction.
2405
01:56:09.760 --> 01:56:14.760
So what is the, from a population-health perspective,
2406
01:56:15.100 --> 01:56:17.000
I think it would be really interesting,
2407
01:56:17.000 --> 01:56:21.810
and this may be a very large collaborative project,
2408
01:56:21.810 --> 01:56:25.340
but in order to look at a systems component
2409
01:56:25.340 --> 01:56:29.380
of both genes and proteins and lipids
2410
01:56:29.380 --> 01:56:33.230
and metabolites that are actually altered along the axes
2411
01:56:33.230 --> 01:56:35.630
either longitudinally or long axes
2412
01:56:35.630 --> 01:56:40.630
of certain processes or severity of addiction,
2413
01:56:42.050 --> 01:56:46.410
or when one is reverting back and being able
2414
01:56:46.410 --> 01:56:50.023
to move away from an addictive a phenotype.
2415
01:56:52.150 --> 01:56:57.150
So when we say to create that full biological narrative,
2416
01:56:58.000 --> 01:57:01.980
the process that Berg undertakes is really
2417
01:57:01.980 --> 01:57:06.980
to commence the, or studies from human tissue samples
2418
01:57:08.511 --> 01:57:12.250
that are clinically annotated,
2419
01:57:12.250 --> 01:57:16.260
ideally these tissue samples are collected over time.
2420
01:57:16.260 --> 01:57:18.130
So we're seeing, we don't believe
2421
01:57:18.130 --> 01:57:20.870
in taking a measurement in one slice of time.
2422
01:57:20.870 --> 01:57:24.940
So looking at these patients in different environments
2423
01:57:24.940 --> 01:57:27.720
over different times in different situations
2424
01:57:27.720 --> 01:57:31.290
and then subjecting these patient samples
2425
01:57:31.290 --> 01:57:36.220
to a very robust multi-omics processing
2426
01:57:36.220 --> 01:57:40.380
that then is then, so the platform is really
2427
01:57:40.380 --> 01:57:44.450
a front-ended biological analysis platform that feeds
2428
01:57:44.450 --> 01:57:48.790
into a back-ended Bayesian AI analytical platform.
2429
01:57:48.790 --> 01:57:53.790
And that then is, it derives in silico insights
2430
01:57:54.710 --> 01:57:56.620
which are then taken back
2431
01:57:56.620 --> 01:57:58.810
into the wet laboratory to be pressure-tested,
2432
01:57:58.810 --> 01:58:01.510
because we feel it's really important
2433
01:58:01.510 --> 01:58:06.510
to come back full circle into the biology.
2434
01:58:06.530 --> 01:58:09.340
So going back into into the wet lab
2435
01:58:10.254 --> 01:58:15.220
and using either CRISPR or RNA, et cetera,
2436
01:58:15.220 --> 01:58:16.990
so that what we're able to do
2437
01:58:16.990 --> 01:58:21.800
is understand how that data generation leads us
2438
01:58:21.800 --> 01:58:25.003
to a certain knowledge topology.
2439
01:58:29.890 --> 01:58:32.700
So I'm not gonna spend tons of time on this
2440
01:58:32.700 --> 01:58:37.700
but because I think this community is very well versed
2441
01:58:38.110 --> 01:58:41.820
on this, but using high-throughput mass spec
2442
01:58:41.820 --> 01:58:46.820
in a LCMS and GCMS platforms to really have a depth
2443
01:58:48.890 --> 01:58:53.440
of understanding with a high quality, high concentration
2444
01:58:53.440 --> 01:58:57.380
of a multi-omics analysis, that then is streamlined
2445
01:58:57.380 --> 01:59:01.660
into really digitizing the biological fingerprints.
2446
01:59:01.660 --> 01:59:02.990
That's what we're really interested
2447
01:59:02.990 --> 01:59:06.590
because you're stacking individual patient omics
2448
01:59:06.590 --> 01:59:09.050
in a manner consistent with understanding
2449
01:59:09.050 --> 01:59:12.940
how to those omics causally infer hypotheses
2450
01:59:12.940 --> 01:59:17.040
within a certain disease, within a certain population,
2451
01:59:17.040 --> 01:59:21.590
within a certain time, what are the, in the case
2452
01:59:21.590 --> 01:59:25.045
of addiction or in a case of Alzheimer's or Parkinson's,
2453
01:59:25.045 --> 01:59:28.780
what is the profile that is consistent
2454
01:59:28.780 --> 01:59:32.680
with a patient feeling L-DOPA or a patient going
2455
01:59:32.680 --> 01:59:36.860
from a mild cognitive impairment
2456
01:59:36.860 --> 01:59:41.370
to full (indistinct) et cetera?
2457
01:59:41.370 --> 01:59:42.473
So these are the types of things
2458
01:59:42.473 --> 01:59:44.370
that at Berg has been focused on
2459
01:59:44.370 --> 01:59:48.840
for the past five to seven years in our neurosciences.
2460
01:59:48.840 --> 01:59:51.700
And from a data-processing perspective,
2461
01:59:51.700 --> 01:59:56.700
it is so, we overuse and prostitute the word data,
2462
01:59:58.100 --> 02:00:01.270
but quality control around data is so important
2463
02:00:01.270 --> 02:00:03.420
and it should not be underestimated.
2464
02:00:03.420 --> 02:00:08.420
We actually have one of our, the head of (indistinct)
2465
02:00:10.124 --> 02:00:12.270
Chief Precision Medicine Officer Dr. Michael Kiebish
2466
02:00:12.270 --> 02:00:14.580
is actually on an IST Advisory Boards.
2467
02:00:14.580 --> 02:00:18.270
And this is where you cannot, especially when it comes
2468
02:00:18.270 --> 02:00:22.560
to brain diseases because the signal-to-noise ratio
2469
02:00:22.560 --> 02:00:27.560
is I think not as well understood as other diseases
2470
02:00:29.900 --> 02:00:32.560
as cancer and diabetes were hallmark mechanisms
2471
02:00:32.560 --> 02:00:35.660
have been around for a longer time.
2472
02:00:35.660 --> 02:00:37.500
So taking this data now and put it
2473
02:00:37.500 --> 02:00:42.160
into a pre-AI analysis enrichment platform
2474
02:00:42.160 --> 02:00:47.160
so that, oh, I should say that the samples and the data
2475
02:00:47.580 --> 02:00:50.880
at Berg, there is not any engagement of priors.
2476
02:00:50.880 --> 02:00:53.900
So we're allowing the patient-driven data
2477
02:00:53.900 --> 02:00:56.150
to drive our hypothesis.
2478
02:00:56.150 --> 02:00:59.620
And after the analysis is done
2479
02:00:59.620 --> 02:01:03.370
on what we refer to internally as basis,
2480
02:01:03.370 --> 02:01:06.993
there is a knowledge apology assessment that's done.
2481
02:01:06.993 --> 02:01:08.670
So based on the inferential causality
2482
02:01:08.670 --> 02:01:13.460
that's predicted in a non-hypothesis
2483
02:01:13.460 --> 02:01:16.060
and data-driven perspective
2484
02:01:16.060 --> 02:01:19.550
then we overlay that into Pubnet, the WIPO,
2485
02:01:19.550 --> 02:01:21.963
the U.S. Patent Office, the Orange Book,
2486
02:01:25.286 --> 02:01:28.190
ClinicalTrials.gov or (indistinct) et cetera,
2487
02:01:28.190 --> 02:01:29.870
just to give you an example.
2488
02:01:29.870 --> 02:01:31.570
And that gives us an understanding
2489
02:01:31.570 --> 02:01:34.560
of what is known about certain genes and proteins.
2490
02:01:34.560 --> 02:01:39.560
If you're seeing a high-confidence networks reproduce
2491
02:01:40.470 --> 02:01:44.910
known drivers of these diseases or addictions,
2492
02:01:44.910 --> 02:01:49.720
it obviously is one a high level of confidence.
2493
02:01:49.720 --> 02:01:52.180
But what we also do, as I said, is to go back
2494
02:01:52.180 --> 02:01:56.140
and perturb that in the wet laboratories
2495
02:01:56.140 --> 02:01:57.763
to validate these outcomes.
2496
02:02:00.360 --> 02:02:04.713
So just, I have a quick video here.
2497
02:02:06.100 --> 02:02:09.570
So just to illustrate this and then mini-animation,
2498
02:02:09.570 --> 02:02:11.510
bringing together these different types
2499
02:02:11.510 --> 02:02:14.250
of omics and different types of data
2500
02:02:14.250 --> 02:02:17.810
as I mentioned from the demographic and clinical data,
2501
02:02:17.810 --> 02:02:21.430
what we allow the Bayesian AI system to do agnostically
2502
02:02:21.430 --> 02:02:26.430
is to really cluster-source and have a concentration
2503
02:02:27.780 --> 02:02:31.410
of certain parameters that then individually
2504
02:02:31.410 --> 02:02:33.570
and by group are subjected
2505
02:02:33.570 --> 02:02:38.350
to the algorithms to construct ensembles of networks.
2506
02:02:38.350 --> 02:02:42.950
And these ensembles of networks really allow us insights
2507
02:02:42.950 --> 02:02:45.520
into high-confidence intervention node.
2508
02:02:45.520 --> 02:02:49.780
Those nodes then give rise to certain control groups,
2509
02:02:49.780 --> 02:02:53.600
whether it's control groups in an unperturbed
2510
02:02:53.600 --> 02:02:57.250
or perturbed state, that allows us to gain insight
2511
02:02:57.250 --> 02:03:00.920
into certain potential drug mechanism of action.
2512
02:03:00.920 --> 02:03:04.960
So we can actually, once our networks are constructed
2513
02:03:04.960 --> 02:03:07.560
to really understand our fundamental basis
2514
02:03:07.560 --> 02:03:10.630
of not only drug-targeting and biomarkers,
2515
02:03:10.630 --> 02:03:14.370
but also (indistinct) so that we can then apply them
2516
02:03:14.370 --> 02:03:16.261
to clinical studies.
2517
02:03:16.261 --> 02:03:18.300
I just, for sake of time I may go a little bit faster here
2518
02:03:18.300 --> 02:03:19.700
on some of the examples.
2519
02:03:19.700 --> 02:03:23.710
I think you all understand how we're doing this,
2520
02:03:23.710 --> 02:03:28.710
but for a specific case study on the oncology,
2521
02:03:29.090 --> 02:03:31.700
we did a very large Alzheimer's solid tumor study
2522
02:03:31.700 --> 02:03:34.082
where we collected tissues and samples
2523
02:03:34.082 --> 02:03:37.200
on patients at baseline every 60 days.
2524
02:03:37.200 --> 02:03:40.573
The FDA allowed us to put
2525
02:03:42.980 --> 02:03:47.420
into the protocol years ago, a multi-omic analysis
2526
02:03:47.420 --> 02:03:51.320
and a Bayesian mapping of individual patients.
2527
02:03:51.320 --> 02:03:53.330
We then were able to understand
2528
02:03:53.330 --> 02:03:56.010
through our mechanism of how a drug work
2529
02:03:56.010 --> 02:03:58.480
from a Warburg-dependent mechanisms
2530
02:03:58.480 --> 02:03:59.800
that we saw better outcomes
2531
02:03:59.800 --> 02:04:03.480
on more highly-metabolic patients, which was not surprising,
2532
02:04:03.480 --> 02:04:05.600
but we pivoted our studies to go
2533
02:04:05.600 --> 02:04:09.690
into phase twos and (indistinct) glioblastoma multiforme
2534
02:04:09.690 --> 02:04:12.800
and pancreatic cancer, both well characterized,
2535
02:04:12.800 --> 02:04:16.220
highly-metabolic tumor environments.
2536
02:04:16.220 --> 02:04:18.180
But what was really interesting is the signatures
2537
02:04:18.180 --> 02:04:22.420
of response and the signatures that correlated
2538
02:04:22.420 --> 02:04:27.420
to a patient who had a Warburg avid tumor
2539
02:04:27.920 --> 02:04:31.100
and had a very aggressive phenotype.
2540
02:04:31.100 --> 02:04:33.330
Those patients who responded to therapy,
2541
02:04:33.330 --> 02:04:38.330
we then did enriched pancreatic cancer population
2542
02:04:38.970 --> 02:04:42.527
in phase two, we used those marker sets
2543
02:04:42.527 --> 02:04:47.320
to confirm if the drug would have continue to work
2544
02:04:47.320 --> 02:04:50.278
with this phenotype, with the imaging data,
2545
02:04:50.278 --> 02:04:52.350
and so it was really this biological readout
2546
02:04:52.350 --> 02:04:54.130
and assessment that was used.
2547
02:04:54.130 --> 02:04:56.747
And we're now pivoting that to go
2548
02:04:56.747 --> 02:05:00.080
into late stage registrational phase three is with respect
2549
02:05:00.080 --> 02:05:03.800
to just the overall ecosystem of systems medicine
2550
02:05:03.800 --> 02:05:05.210
to deconstruct depression
2551
02:05:05.210 --> 02:05:07.610
and neurological cognitive functions.
2552
02:05:07.610 --> 02:05:10.110
We've really prided ourselves on working
2553
02:05:10.110 --> 02:05:15.110
with many of you, leading academics and other companies
2554
02:05:15.760 --> 02:05:20.400
with NIH, some specific laboratories in NIH,
2555
02:05:20.400 --> 02:05:24.517
and really seek to understand different impact factors.
2556
02:05:24.517 --> 02:05:27.520
And we do have a neurodegenerative program that's focused
2557
02:05:27.520 --> 02:05:30.100
on Parkinson's and Alzheimer's, but looking
2558
02:05:30.100 --> 02:05:34.650
at the systems functions from the metabolomics
2559
02:05:34.650 --> 02:05:37.650
and lipidomics and how those axes change
2560
02:05:37.650 --> 02:05:40.220
in different brain diseases
2561
02:05:40.220 --> 02:05:45.220
and being able to use the multi-omic signatures
2562
02:05:47.057 --> 02:05:51.540
as a glue, towards constructing these networks
2563
02:05:51.540 --> 02:05:54.360
so then we generate hypothesis to be tested
2564
02:05:54.360 --> 02:05:59.280
along the axes of bio-marker and drug development.
2565
02:05:59.280 --> 02:06:01.570
So with respect to substance abuse
2566
02:06:01.570 --> 02:06:05.143
being such a hallmark multifactorial challenge,
2567
02:06:06.060 --> 02:06:08.480
the approach that we're taking
2568
02:06:08.480 --> 02:06:13.480
by the upfront biology and the omics screening
2569
02:06:13.910 --> 02:06:17.353
really does allow us to work
2570
02:06:17.353 --> 02:06:19.540
with some of the leading labs
2571
02:06:19.540 --> 02:06:24.540
to study trauma as a, or from a mental health perspective
2572
02:06:25.981 --> 02:06:30.100
in the environmental impact factors that span
2573
02:06:30.100 --> 02:06:34.010
the micro-constructs of economics and access
2574
02:06:34.010 --> 02:06:38.900
and behavioral mechanisms and et cetera, et cetera,
2575
02:06:38.900 --> 02:06:41.580
the fundamental biological readouts
2576
02:06:41.580 --> 02:06:42.980
and mental health illnesses,
2577
02:06:43.833 --> 02:06:46.360
and really integrating this real-world data
2578
02:06:46.360 --> 02:06:50.100
and molecular phenotypes into a data-driven understanding
2579
02:06:50.100 --> 02:06:52.080
and predictive components
2580
02:06:52.080 --> 02:06:56.404
so that we can then further understand what are some
2581
02:06:56.404 --> 02:06:57.237
of these driving factors
2582
02:06:57.237 --> 02:06:59.820
and crowdsource that validation
2583
02:06:59.820 --> 02:07:02.830
with the hundreds of individuals
2584
02:07:02.830 --> 02:07:05.542
that are that are on the Zoom right now.
2585
02:07:05.542 --> 02:07:09.030
What are some predisposing biomarkers?
2586
02:07:09.030 --> 02:07:12.480
Can we start to do that at a preventative component?
2587
02:07:12.480 --> 02:07:16.540
Can we do it where we identify early risk factors
2588
02:07:16.540 --> 02:07:21.540
in very young individuals
2589
02:07:21.640 --> 02:07:23.270
so we have a fundamental understanding
2590
02:07:23.270 --> 02:07:27.750
of maybe from an actuary sciences perspective?
2591
02:07:27.750 --> 02:07:30.030
And then support the point-of-care decisions
2592
02:07:30.030 --> 02:07:32.290
with the physicians and the caregivers
2593
02:07:32.290 --> 02:07:37.290
and individuals in, or health experts
2594
02:07:38.190 --> 02:07:40.120
in rehab centers, et cetera,
2595
02:07:40.120 --> 02:07:44.210
so that this ecosystem of stakeholders within the family,
2596
02:07:44.210 --> 02:07:47.730
the patient, the insurance companies, hospital systems,
2597
02:07:47.730 --> 02:07:51.010
their employers, their environment
2598
02:07:51.010 --> 02:07:52.580
of these individuals, because I think this
2599
02:07:52.580 --> 02:07:56.720
is a, I know I'm preaching to the choir here
2600
02:07:56.720 --> 02:08:01.720
but this is a community-based issue for society,
2601
02:08:01.880 --> 02:08:05.790
and we have all as scientists and physicians
2602
02:08:05.790 --> 02:08:10.290
a moral responsibility to bring these disparate data sets
2603
02:08:10.290 --> 02:08:13.763
and the diversity of thought constructs into our studies.
2604
02:08:17.100 --> 02:08:20.440
And as an exemplar of that, we believe
2605
02:08:20.440 --> 02:08:25.440
in both biological, but also global inclusion of diversity.
2606
02:08:26.690 --> 02:08:28.250
What are some of the factors
2607
02:08:28.250 --> 02:08:32.290
that whether it's urban or suburban or international
2608
02:08:32.290 --> 02:08:35.925
or certain predisposing factors of stresses
2609
02:08:35.925 --> 02:08:40.925
that somebody in Cambridge, Massachusetts,
2610
02:08:41.610 --> 02:08:46.610
may not ever understand to bring that to bear fundamentally
2611
02:08:47.170 --> 02:08:48.993
within our studies?
2612
02:08:49.956 --> 02:08:50.961
And then we-
Niven,
2613
02:08:50.961 --> 02:08:51.990
we're just about of time.
2614
02:08:51.990 --> 02:08:53.780
Yes, and I just wanna say thank you
2615
02:08:53.780 --> 02:08:57.740
to the patients and families that have participated,
2616
02:08:57.740 --> 02:09:00.133
and thank you for the opportunity to present.
2617
02:09:06.250 --> 02:09:07.963
Thank you, that was great, Dr. Narain.
2618
02:09:08.960 --> 02:09:11.140
Just a reminder to everybody, we will have a Q&A session
2619
02:09:11.140 --> 02:09:14.230
at the end after all the speakers go.
2620
02:09:14.230 --> 02:09:17.100
So our next speaker is Dr. Olivier Elemento
2621
02:09:17.100 --> 02:09:18.690
from Weill Cornell Medicine
2622
02:09:18.690 --> 02:09:20.370
where he is at tenured full professor
2623
02:09:20.370 --> 02:09:22.500
of their Institute for Precision Medicine.
2624
02:09:22.500 --> 02:09:24.290
Dr. Elemento has led the development
2625
02:09:24.290 --> 02:09:26.400
of novel clinical genomics assays
2626
02:09:26.400 --> 02:09:28.670
and is currently leading a large multi-disease effort
2627
02:09:28.670 --> 02:09:31.840
to bring whole genome sequencing into clinical practice.
2628
02:09:31.840 --> 02:09:33.770
He has received many honors, has co-founded
2629
02:09:33.770 --> 02:09:36.260
two venture-capital-funded companies and serves
2630
02:09:36.260 --> 02:09:37.640
on the scientific advisory boards
2631
02:09:37.640 --> 02:09:39.100
of several other companies.
2632
02:09:39.100 --> 02:09:41.300
The title of his talk today is Predicting Mechanisms
2633
02:09:41.300 --> 02:09:43.470
of Action, Human Toxicity
2634
02:09:43.470 --> 02:09:46.113
and Optimizing Indication Using Biology-driven AI.
2635
02:09:49.022 --> 02:09:50.025
Thank you Susan.
2636
02:09:50.025 --> 02:09:50.858
I just wanna make sure
2637
02:09:50.858 --> 02:09:52.580
that you hear me well and that you see my slides.
2638
02:09:52.580 --> 02:09:53.820
Yes to both.
2639
02:09:53.820 --> 02:09:54.740
Great, fantastic.
2640
02:09:54.740 --> 02:09:55.840
Well, thanks so much.
2641
02:09:55.840 --> 02:10:00.450
It's a wonderful honor and pleasure to be here today.
2642
02:10:00.450 --> 02:10:01.640
I wanna start with a disclosure,
2643
02:10:01.640 --> 02:10:05.940
which is that obviously having heard some of the talks,
2644
02:10:05.940 --> 02:10:09.300
as you'll see the content of my talk is a bit different.
2645
02:10:09.300 --> 02:10:10.558
Our main research focus
2646
02:10:10.558 --> 02:10:15.558
is primarily cancer and genomics and precision medicine.
2647
02:10:15.900 --> 02:10:19.140
We're actually starting to do work in addiction, looking
2648
02:10:19.140 --> 02:10:22.980
for biomarkers of addiction and looking for maybe using some
2649
02:10:22.980 --> 02:10:25.240
of the techniques that I'm gonna talk about today
2650
02:10:25.240 --> 02:10:29.090
to identify novel molecules that could be used
2651
02:10:29.090 --> 02:10:30.210
in the context of addiction,
2652
02:10:30.210 --> 02:10:33.560
but it's not necessarily the main focus of my research.
2653
02:10:33.560 --> 02:10:35.640
Although as I say, I do think
2654
02:10:35.640 --> 02:10:37.050
that some of what I'm going to talk about today
2655
02:10:37.050 --> 02:10:38.513
has broad application,
2656
02:10:39.995 --> 02:10:41.897
and hopefully you will be able to discuss this
2657
02:10:41.897 --> 02:10:43.323
after the talk.
2658
02:10:44.570 --> 02:10:47.610
So a couple of disclosures, first, as I mentioned,
2659
02:10:47.610 --> 02:10:51.660
I did co-found two companies that do work,
2660
02:10:51.660 --> 02:10:53.770
some of them at least like one free biotech
2661
02:10:53.770 --> 02:10:57.126
in the field of AI-driven drug discovery,
2662
02:10:57.126 --> 02:10:59.810
again, a bit more focused on oncology,
2663
02:10:59.810 --> 02:11:02.920
but I'm also on the new board of several companies
2664
02:11:02.920 --> 02:11:06.890
and getting funding from multiple pharma companies
2665
02:11:06.890 --> 02:11:08.040
as part of my research.
2666
02:11:09.520 --> 02:11:13.460
At Cornell, my group is a combined computational biology
2667
02:11:13.460 --> 02:11:16.390
and bench biology lab.
2668
02:11:16.390 --> 02:11:18.620
So we do a bit of both.
2669
02:11:18.620 --> 02:11:20.500
We've been using AI for a variety
2670
02:11:20.500 --> 02:11:23.660
of different projects now in my group.
2671
02:11:23.660 --> 02:11:25.570
And I'm excited today to tell you
2672
02:11:25.570 --> 02:11:27.350
about at least some of the applications of AI
2673
02:11:27.350 --> 02:11:30.745
that we've been sort of using.
2674
02:11:30.745 --> 02:11:33.360
As I said, this is a whole range
2675
02:11:33.360 --> 02:11:35.718
of applications that you see on the screen here
2676
02:11:35.718 --> 02:11:37.280
and certainly be happy to come back and give you
2677
02:11:37.280 --> 02:11:40.840
more information about for example, how we can predict
2678
02:11:40.840 --> 02:11:45.030
which combinations of molecules are synergistic altogether,
2679
02:11:45.030 --> 02:11:46.509
which I think
2680
02:11:46.509 --> 02:11:49.290
is a very important problem to address (clears throat)
2681
02:11:49.290 --> 02:11:52.920
But today I'm going to focus on two aspects of our research.
2682
02:11:52.920 --> 02:11:56.050
One is the prediction of clinical trials outcomes
2683
02:11:56.050 --> 02:11:58.360
and prediction of toxicity of molecules
2684
02:11:58.360 --> 02:12:01.160
before we actually give those molecules to human beings.
2685
02:12:02.183 --> 02:12:03.790
And the second is about predicting the targets
2686
02:12:03.790 --> 02:12:06.540
of small molecules that do not have a known mechanism
2687
02:12:06.540 --> 02:12:10.880
of action, but you know what, who's identification
2688
02:12:10.880 --> 02:12:13.280
informs clinical trials as you'll see
2689
02:12:13.280 --> 02:12:14.603
in a very positive way.
2690
02:12:15.478 --> 02:12:17.420
So I'm gonna start by research that we've started
2691
02:12:17.420 --> 02:12:21.260
a few years ago now where we've been able to build
2692
02:12:21.260 --> 02:12:25.100
predictive models of toxicity of small molecules in humans.
2693
02:12:25.100 --> 02:12:29.180
The idea here was really to learn from a very large number
2694
02:12:29.180 --> 02:12:32.790
of clinical trials that have already been done
2695
02:12:32.790 --> 02:12:35.370
to differentiate between trials that failed
2696
02:12:35.370 --> 02:12:38.840
for reasons of toxicity, kind of broadly,
2697
02:12:38.840 --> 02:12:40.300
and then specifically focusing
2698
02:12:40.300 --> 02:12:42.930
on specific toxicity mechanisms
2699
02:12:42.930 --> 02:12:46.560
and trials that went all the way to FDA approval.
2700
02:12:46.560 --> 02:12:48.777
So we've been trying to identify features
2701
02:12:48.777 --> 02:12:53.070
of the molecules as well as the targets of the molecules.
2702
02:12:53.070 --> 02:12:54.320
And I think combining
2703
02:12:54.320 --> 02:12:56.143
that in the order of about 60,
2704
02:12:57.064 --> 02:12:59.393
65 features, again, really, a broad range
2705
02:12:59.393 --> 02:13:04.393
of features focused on both the drug and the target
2706
02:13:04.858 --> 02:13:07.170
and what the target is doing in normal tissue.
2707
02:13:07.170 --> 02:13:09.750
And the idea has been to use machine learning
2708
02:13:09.750 --> 02:13:11.890
as a way to integrate all these different pieces
2709
02:13:11.890 --> 02:13:16.890
of information and predict which trials sort of failed
2710
02:13:17.810 --> 02:13:20.363
in phase one because of toxicity.
2711
02:13:21.440 --> 02:13:22.780
And so we've actually had quite a bit
2712
02:13:22.780 --> 02:13:24.243
of success in doing so,
2713
02:13:25.485 --> 02:13:28.330
again, the integration of these different data types
2714
02:13:28.330 --> 02:13:30.630
and modalities I think has been very effective
2715
02:13:30.630 --> 02:13:32.993
in terms of predicting outcomes of trials.
2716
02:13:34.260 --> 02:13:37.060
This is here kind of a score that we get
2717
02:13:37.060 --> 02:13:39.041
from the predictive models.
2718
02:13:39.041 --> 02:13:40.640
We've been looking at different types of predictive models.
2719
02:13:40.640 --> 02:13:42.750
The one that we've used
2720
02:13:42.750 --> 02:13:44.070
in this particular context
2721
02:13:44.070 --> 02:13:47.400
has been the random forest sort of type of model,
2722
02:13:47.400 --> 02:13:49.749
which is something that we like it's not kind
2723
02:13:49.749 --> 02:13:52.550
of as trendy, as deep learning as some
2724
02:13:52.550 --> 02:13:56.970
of the more neural network models that are being used now.
2725
02:13:56.970 --> 02:13:59.660
But it is good for a variety of different reasons,
2726
02:13:59.660 --> 02:14:02.070
one of them is that it can integrate for diverse data types.
2727
02:14:02.070 --> 02:14:03.227
And that's something that is important
2728
02:14:03.227 --> 02:14:06.400
in the real world to be able to integrate
2729
02:14:06.400 --> 02:14:09.480
by (indistinct) data types, continuous data types
2730
02:14:09.480 --> 02:14:11.950
and so on, and not paying attention too much
2731
02:14:11.950 --> 02:14:14.750
about the distribution of feature values and so on.
2732
02:14:14.750 --> 02:14:17.807
So anyway, so the point is really that the score that we get
2733
02:14:17.807 --> 02:14:20.500
is equal to probability of approval that we get
2734
02:14:20.500 --> 02:14:22.680
out of this kind of machine learning model
2735
02:14:22.680 --> 02:14:26.200
is quite effective at finding the molecules that failed
2736
02:14:26.200 --> 02:14:28.240
clinical trials in the past.
2737
02:14:28.240 --> 02:14:30.070
In fact, it's actually a lot more effective,
2738
02:14:30.070 --> 02:14:31.470
what you see on the right side here,
2739
02:14:31.470 --> 02:14:33.900
than some of the rules that have been used
2740
02:14:33.900 --> 02:14:36.660
in industry to predict what a drug-like molecule
2741
02:14:36.660 --> 02:14:38.520
looks like or should look like,
2742
02:14:38.520 --> 02:14:41.770
like rule of five (indistinct)
2743
02:14:41.770 --> 02:14:44.060
We see that these measures
2744
02:14:44.060 --> 02:14:46.760
which are widely used really failed
2745
02:14:46.760 --> 02:14:50.010
at predicting which molecules are gonna be toxic
2746
02:14:50.010 --> 02:14:54.000
in trials while the machine learning model that we built,
2747
02:14:54.000 --> 02:14:54.833
which you see here
2748
02:14:54.833 --> 02:14:58.290
in black, is actually quite effective at very predictive.
2749
02:14:58.290 --> 02:14:59.590
So I think it's really a testament
2750
02:14:59.590 --> 02:15:02.330
to using a data science approach,
2751
02:15:02.330 --> 02:15:05.590
learning from data to essentially achieve
2752
02:15:05.590 --> 02:15:10.590
a particular outcome that is in this particular case
2753
02:15:10.630 --> 02:15:12.553
predicting toxicity of molecules.
2754
02:15:13.430 --> 02:15:16.390
So we were able to predict toxicity across the board,
2755
02:15:16.390 --> 02:15:18.480
basically failure of clinical trial,
2756
02:15:18.480 --> 02:15:21.310
but we also build models that are very specific
2757
02:15:21.310 --> 02:15:24.573
to a particular type of toxicity, for example, neutropenia,
2758
02:15:25.857 --> 02:15:27.480
nephropathy or pancreatitis and so on.
2759
02:15:27.480 --> 02:15:30.480
As you can see here based on the AUC levels
2760
02:15:30.480 --> 02:15:32.500
and accuracy levels that we achieve,
2761
02:15:32.500 --> 02:15:34.880
we can build models that are very predictive.
2762
02:15:34.880 --> 02:15:37.720
I think it's really kinda cool to think about again,
2763
02:15:37.720 --> 02:15:39.860
using this kind of data science approach learning
2764
02:15:39.860 --> 02:15:43.970
from large number of trials and really building models
2765
02:15:43.970 --> 02:15:45.883
that sometimes they may not be complex,
2766
02:15:45.883 --> 02:15:48.590
that integrate multiple different rules
2767
02:15:48.590 --> 02:15:50.750
but are quite predictive.
2768
02:15:50.750 --> 02:15:53.390
This platform was able, for example, to pull,
2769
02:15:53.390 --> 02:15:57.010
identify molecules that were initially approved
2770
02:15:57.010 --> 02:15:58.180
but had to be withdrawn
2771
02:15:58.180 --> 02:16:00.800
because of toxicity and side effects.
2772
02:16:00.800 --> 02:16:02.370
One of them was Avandia for example,
2773
02:16:02.370 --> 02:16:05.013
which is a molecule that had to be withdrawn,
2774
02:16:06.530 --> 02:16:09.600
and method was able to predict
2775
02:16:09.600 --> 02:16:12.230
that indeed this molecule was gonna be toxic
2776
02:16:12.230 --> 02:16:13.360
in human beings.
2777
02:16:13.360 --> 02:16:16.040
So just again, examples of utilization
2778
02:16:16.040 --> 02:16:17.690
of this kind of predictive models
2779
02:16:18.740 --> 02:16:21.146
for basically avoiding
2780
02:16:21.146 --> 02:16:24.383
this kind of outcome here in the future.
2781
02:16:25.340 --> 02:16:27.979
So switching gear, so the other application
2782
02:16:27.979 --> 02:16:30.120
that I'm gonna talk about today is the idea that you can use
2783
02:16:30.120 --> 02:16:31.480
AI, you can use machine learning
2784
02:16:31.480 --> 02:16:36.480
as a way to maybe identify mechanisms of actions
2785
02:16:36.510 --> 02:16:40.010
for molecules that don't have a mechanism of action
2786
02:16:40.979 --> 02:16:43.909
and therefore can't really advance too well in the kind
2787
02:16:43.909 --> 02:16:47.530
of the drug development sort of process.
2788
02:16:48.390 --> 02:16:50.680
A lot of these molecules exist.
2789
02:16:50.680 --> 02:16:53.670
Typically they come from phenotypic screens
2790
02:16:53.670 --> 02:16:56.001
or phenotypic screening processes,
2791
02:16:56.001 --> 02:16:58.610
molecules and have some efficacy let's say,
2792
02:16:58.610 --> 02:17:00.910
in cell lines or even PDX models,
2793
02:17:00.910 --> 02:17:03.920
but because they don't have one understood mechanism
2794
02:17:03.920 --> 02:17:07.710
of action, it's very hard to design clinical trials
2795
02:17:07.710 --> 02:17:11.230
to advance themselves and then really identify
2796
02:17:11.230 --> 02:17:14.050
which patient populations may benefit the most
2797
02:17:14.050 --> 02:17:15.920
from these molecules.
2798
02:17:15.920 --> 02:17:18.680
So this is something that we've been working on
2799
02:17:18.680 --> 02:17:19.770
for some time.
2800
02:17:19.770 --> 02:17:23.364
We've been using experimental methods to find a target
2801
02:17:23.364 --> 02:17:24.197
of some molecules.
2802
02:17:24.197 --> 02:17:27.317
This is an attempt here to use AI to do so.
2803
02:17:27.317 --> 02:17:29.370
And as you'll see, a successful attempt,
2804
02:17:29.370 --> 02:17:32.700
and this is what that's been be happening in my group
2805
02:17:32.700 --> 02:17:35.590
by Dr. Madhukar, who's on the slide here,
2806
02:17:35.590 --> 02:17:39.000
and who since started one free biotech company
2807
02:17:39.000 --> 02:17:40.260
that has been under my lab.
2808
02:17:40.260 --> 02:17:43.190
So the challenge really here
2809
02:17:43.190 --> 02:17:45.809
is that for molecules that don't have a target,
2810
02:17:45.809 --> 02:17:47.150
and here we can have lots of different targets
2811
02:17:47.150 --> 02:17:50.180
but lots of proteins that these molecules could target.
2812
02:17:50.180 --> 02:17:51.650
In fact sometimes
2813
02:17:51.650 --> 02:17:53.250
a molecule can target multiple proteins.
2814
02:17:53.250 --> 02:17:57.370
So the identification of the target of a small molecule
2815
02:17:57.370 --> 02:18:01.220
is very difficult given the complexity of a task.
2816
02:18:01.220 --> 02:18:05.038
What we've done is to build a Bayesian method
2817
02:18:05.038 --> 02:18:07.730
to find the target, so to predict the target
2818
02:18:07.730 --> 02:18:09.130
of small molecules.
2819
02:18:09.130 --> 02:18:10.670
The idea here is really to learn
2820
02:18:10.670 --> 02:18:14.770
from many, many, many targets or many molecules
2821
02:18:14.770 --> 02:18:17.460
for which we know the target, how do we can connect
2822
02:18:17.460 --> 02:18:20.320
the molecule and the target.
2823
02:18:20.320 --> 02:18:22.470
And this is not done in absence of data.
2824
02:18:22.470 --> 02:18:25.570
Of course, this is done in the presence
2825
02:18:25.570 --> 02:18:29.080
of what we think is reasonably sort of accessible type
2826
02:18:29.080 --> 02:18:31.150
of data, which one at a time
2827
02:18:31.150 --> 02:18:34.400
doesn't really give you the target that can gives you clues,
2828
02:18:34.400 --> 02:18:37.626
but what we showed in this in this work here
2829
02:18:37.626 --> 02:18:39.540
is that when you add up these clues,
2830
02:18:39.540 --> 02:18:41.891
you basically kind of narrow down to the target
2831
02:18:41.891 --> 02:18:42.800
of small molecules really well.
2832
02:18:42.800 --> 02:18:45.380
So one of the clues that is obviously very effective
2833
02:18:45.380 --> 02:18:50.280
is the target, sorry, the structure of the molecule itself.
2834
02:18:50.280 --> 02:18:52.780
It does tell you a lot about potential targets
2835
02:18:52.780 --> 02:18:56.610
because we have a universe of all possible molecules.
2836
02:18:56.610 --> 02:18:58.230
Some molecules have a known target,
2837
02:18:58.230 --> 02:19:00.360
so we can place a small molecule
2838
02:19:00.360 --> 02:19:02.020
in the universe of these targets
2839
02:19:02.020 --> 02:19:03.600
and therefore somehow guess.
2840
02:19:03.600 --> 02:19:06.310
But it's not a very robust way to do this.
2841
02:19:06.310 --> 02:19:08.940
There's also information about expression levels,
2842
02:19:08.940 --> 02:19:10.310
things that happen to cells.
2843
02:19:10.310 --> 02:19:11.690
So cell lines, when you treat them
2844
02:19:11.690 --> 02:19:13.830
with a particular small molecule,
2845
02:19:13.830 --> 02:19:16.890
genes that go up and down, that doesn't really even tell you
2846
02:19:16.890 --> 02:19:19.820
the target but it does give you some clues,
2847
02:19:19.820 --> 02:19:21.260
viability of cell lines,
2848
02:19:21.260 --> 02:19:23.682
the panel of cell lines that you treat
2849
02:19:23.682 --> 02:19:25.750
with the same small molecule, some cell lines die,
2850
02:19:25.750 --> 02:19:28.050
some cell lines don't (indistinct)
2851
02:19:28.050 --> 02:19:31.460
That also carries information about the target.
2852
02:19:31.460 --> 02:19:34.080
And last but not least, side effect profiles
2853
02:19:34.080 --> 02:19:35.250
that you get
2854
02:19:35.250 --> 02:19:37.320
from animals likewise do when you have them
2855
02:19:37.320 --> 02:19:39.390
probably give you information about the target.
2856
02:19:39.390 --> 02:19:41.702
So the point about this whole slide here
2857
02:19:41.702 --> 02:19:43.692
is that we were able to put together all
2858
02:19:43.692 --> 02:19:47.010
of these bits of information and integrate
2859
02:19:47.010 --> 02:19:50.370
these data types using a Bayesian framework,
2860
02:19:50.370 --> 02:19:52.010
and we showed using this framework
2861
02:19:52.010 --> 02:19:54.819
that the more information you have, the more
2862
02:19:54.819 --> 02:19:56.700
of these data types that you have on the left side,
2863
02:19:56.700 --> 02:19:59.250
the more you're able to predict the target
2864
02:19:59.250 --> 02:20:00.360
of a small molecule.
2865
02:20:00.360 --> 02:20:02.788
But I think most importantly
2866
02:20:02.788 --> 02:20:05.477
as you can see here, our ability to predict the target
2867
02:20:05.477 --> 02:20:08.177
of the small molecule when we have quite a bit of data
2868
02:20:09.410 --> 02:20:10.870
as you see here is actually quite high.
2869
02:20:10.870 --> 02:20:13.910
And we're talking about an AUC of about 0.9,
2870
02:20:13.910 --> 02:20:15.860
which is somewhere around 90% accuracy,
2871
02:20:17.260 --> 02:20:18.540
which is quite remarkable.
2872
02:20:18.540 --> 02:20:20.230
And this has done in the blind search.
2873
02:20:20.230 --> 02:20:23.621
This is really kind of done in a way that we mix
2874
02:20:23.621 --> 02:20:26.300
the real-life situation.
2875
02:20:26.300 --> 02:20:30.137
So the idea of his method is really that we end up
2876
02:20:30.137 --> 02:20:33.020
adding up these clues and being able to position
2877
02:20:33.020 --> 02:20:33.980
these new molecules
2878
02:20:33.980 --> 02:20:36.783
in the context of this universe of molecules
2879
02:20:36.783 --> 02:20:38.160
that have a known target,
2880
02:20:38.160 --> 02:20:41.477
and that's how we effectively guess
2881
02:20:41.477 --> 02:20:44.830
the target of a molecule that does not have a target.
2882
02:20:44.830 --> 02:20:45.840
We are able to make a lot
2883
02:20:45.840 --> 02:20:47.330
of different really interesting connections
2884
02:20:47.330 --> 02:20:50.250
by doing this, pretty relevant to this particular meeting.
2885
02:20:50.250 --> 02:20:53.355
For example, we've found that certain drugs
2886
02:20:53.355 --> 02:20:57.990
that have a reasonably well known mechanism of action
2887
02:20:57.990 --> 02:20:59.690
do kind of cluster in this kind
2888
02:20:59.690 --> 02:21:02.440
of functional space with all the drugs that have
2889
02:21:02.440 --> 02:21:03.890
a completely different mechanism of action.
2890
02:21:03.890 --> 02:21:07.170
For example codeine is clustering pretty close to paclitaxel
2891
02:21:07.170 --> 02:21:11.690
which is a taxane inhibitor, which I think is a clue
2892
02:21:11.690 --> 02:21:13.856
that codeine for example could have
2893
02:21:13.856 --> 02:21:16.520
kind of a microtubule inhibiting
2894
02:21:16.520 --> 02:21:18.510
sort of type of function.
2895
02:21:18.510 --> 02:21:20.480
And in fact, this is something that we tested.
2896
02:21:20.480 --> 02:21:22.990
I don't have data here, but we tested that hypothesis
2897
02:21:22.990 --> 02:21:24.740
and actually really that did come up
2898
02:21:24.740 --> 02:21:29.740
as a mechanism of activity of codeine.
2899
02:21:30.080 --> 02:21:32.210
So this is something that obviously it would have
2900
02:21:32.210 --> 02:21:34.360
to be followed up sort of solve that.
2901
02:21:35.330 --> 02:21:37.927
The application of ease analysis to a lot
2902
02:21:37.927 --> 02:21:40.650
of molecules that did not have as a target was show
2903
02:21:40.650 --> 02:21:44.113
that we can predict reliably targets in about 40% of cases.
2904
02:21:45.030 --> 02:21:48.300
We showed that a lot of these molecules
2905
02:21:48.300 --> 02:21:51.260
that you did not have a target before, we were interested
2906
02:21:51.260 --> 02:21:55.430
in specific classes, one of them was microtubule inhibition.
2907
02:21:55.430 --> 02:21:56.920
We actually did find
2908
02:21:56.920 --> 02:21:59.700
quite a few novel microtubule inhibitors according
2909
02:21:59.700 --> 02:22:01.750
to our prediction.
2910
02:22:01.750 --> 02:22:02.930
We tested those predictions
2911
02:22:02.930 --> 02:22:06.270
in the lab and found that out of 24 predictions
2912
02:22:06.270 --> 02:22:09.140
of small molecules that known to predict
2913
02:22:09.140 --> 02:22:12.080
or to be predicted to find microtubules,
2914
02:22:12.080 --> 02:22:16.160
14 of them actually were affecting microtubules, sorry.
2915
02:22:16.160 --> 02:22:19.050
Yeah, so this is again kind of a nice validation
2916
02:22:19.050 --> 02:22:21.060
of the approach.
2917
02:22:21.060 --> 02:22:23.120
And in fact, some of these molecules were actually able
2918
02:22:23.120 --> 02:22:26.940
to reverse resistance to known microtubule inhibitors.
2919
02:22:26.940 --> 02:22:29.370
Not only we found novel microtubule inhibitors,
2920
02:22:29.370 --> 02:22:30.870
but some of them have novel function
2921
02:22:30.870 --> 02:22:34.340
that is probably connected to a different mechanism
2922
02:22:34.340 --> 02:22:36.660
of inhibition and microtubules
2923
02:22:37.583 --> 02:22:39.270
and it's a different function.
2924
02:22:39.270 --> 02:22:41.050
So another analysis that we did
2925
02:22:41.050 --> 02:22:43.860
was to apply this to kind of, again, a real-life scenario.
2926
02:22:43.860 --> 02:22:47.070
We worked with a company out of Philadelphia that had
2927
02:22:47.070 --> 02:22:48.270
a molecule called ONC201
2928
02:22:49.218 --> 02:22:52.870
that was in the data was able to track cancer cells
2929
02:22:52.870 --> 02:22:54.960
but did not have a known mechanism of action.
2930
02:22:54.960 --> 02:22:57.210
It came out of affinitive screen.
2931
02:22:57.210 --> 02:22:59.997
You really, the the company was struggling to find
2932
02:22:59.997 --> 02:23:03.180
the target of this molecule, and as a result,
2933
02:23:03.180 --> 02:23:06.220
was not able to have successful clinical trials.
2934
02:23:06.220 --> 02:23:07.720
They went all the way to clinical trials
2935
02:23:07.720 --> 02:23:12.080
because this molecule has a lot of efficacy
2936
02:23:12.080 --> 02:23:14.410
in a very small subset of patients,
2937
02:23:14.410 --> 02:23:17.507
but they were never able to understand what were the subset
2938
02:23:17.507 --> 02:23:20.040
and how to prioritize them.
2939
02:23:20.040 --> 02:23:22.470
So we came in and used this method that I showed you,
2940
02:23:22.470 --> 02:23:24.486
exactly the same method.
2941
02:23:24.486 --> 02:23:25.319
They had a bit of data about the molecule
2942
02:23:25.319 --> 02:23:28.780
and were able to plug that into our Bayesian Framework.
2943
02:23:28.780 --> 02:23:32.970
And we found, but very unexpectedly actually, the molecule
2944
02:23:32.970 --> 02:23:35.490
was predicted to target a dopamine receptor,
2945
02:23:35.490 --> 02:23:39.060
which is quite really not a a cancer target in the fields
2946
02:23:39.060 --> 02:23:40.670
really something that's quite unexpected
2947
02:23:40.670 --> 02:23:42.300
as a prediction.
2948
02:23:42.300 --> 02:23:44.640
To the credit to the company, they took a leap of faith
2949
02:23:44.640 --> 02:23:47.830
and went back to the lab and did all kinds
2950
02:23:47.830 --> 02:23:51.290
of additional experiments to demonstrate that indeed
2951
02:23:51.290 --> 02:23:52.870
this molecule is able to bind
2952
02:23:52.870 --> 02:23:56.170
to to dopamine receptors and alter its function.
2953
02:23:56.170 --> 02:23:57.810
And so I'm not gonna go into details
2954
02:23:57.810 --> 02:24:00.800
but these are all kinds of different assays, binding assays,
2955
02:24:00.800 --> 02:24:03.810
kind of functional assays, like TBT essays, as well
2956
02:24:03.810 --> 02:24:06.850
as looking at patient samples and what happens to them
2957
02:24:06.850 --> 02:24:09.683
after treatment of this molecule, and they did finding
2958
02:24:09.683 --> 02:24:11.999
that the downstream effect of the molecule
2959
02:24:11.999 --> 02:24:15.680
is really there as illustrated, for example,
2960
02:24:15.680 --> 02:24:17.850
for by this prolactin expression.
2961
02:24:17.850 --> 02:24:20.760
So the point of all this is that DRD2
2962
02:24:20.760 --> 02:24:24.300
really was physically shown and proven to be bound
2963
02:24:24.300 --> 02:24:26.071
and targeted by ONC201.
2964
02:24:26.071 --> 02:24:27.970
And that's, if our company was a game changer
2965
02:24:27.970 --> 02:24:29.580
because it opened up the possibility
2966
02:24:29.580 --> 02:24:33.960
of designing trials that were specifically focused
2967
02:24:33.960 --> 02:24:37.990
on tumors and disease that express high level of the target,
2968
02:24:37.990 --> 02:24:40.070
one of them was gliomas,
2969
02:24:40.070 --> 02:24:43.280
especially H3K27M mutated gliomas,
2970
02:24:43.280 --> 02:24:46.640
and really long story short, the trials told us
2971
02:24:47.615 --> 02:24:49.370
and has been a major success.
2972
02:24:49.370 --> 02:24:53.910
Really this is a very dire disease typically
2973
02:24:53.910 --> 02:24:55.080
that has very little survival
2974
02:24:55.080 --> 02:24:56.933
of your disease in the short-term.
2975
02:24:58.080 --> 02:25:01.130
This trial here had actually quite a few patients
2976
02:25:01.130 --> 02:25:04.470
who did extremely well and was probably a size as a result.
2977
02:25:04.470 --> 02:25:05.770
And it's actually a create a lot
2978
02:25:05.770 --> 02:25:08.500
of excitement in the neuro-oncology community
2979
02:25:08.500 --> 02:25:11.660
as a result of this efficacy.
2980
02:25:11.660 --> 02:25:13.280
So again, this is a great example
2981
02:25:13.280 --> 02:25:16.390
of how a molecule that was had no target was able
2982
02:25:16.390 --> 02:25:19.810
to be prioritized according to an AI method
2983
02:25:19.810 --> 02:25:22.710
in a way that really wasn't possible before.
2984
02:25:22.710 --> 02:25:24.890
Just one last side, we are able
2985
02:25:24.890 --> 02:25:27.430
to now to identify additional two more types
2986
02:25:27.430 --> 02:25:30.240
where this molecule has potential additional efficacy.
2987
02:25:30.240 --> 02:25:32.840
One of those is metastatic prostate cancer.
2988
02:25:32.840 --> 02:25:34.710
We found high expression of DRD2
2989
02:25:34.710 --> 02:25:36.827
in those using (indistinct)
2990
02:25:36.827 --> 02:25:37.660
As you can see in the middle here,
2991
02:25:37.660 --> 02:25:40.160
we're able to show that indeed this molecule is effective
2992
02:25:40.160 --> 02:25:45.030
at killing these prostate cancer cells.
2993
02:25:45.030 --> 02:25:46.460
And again, this is something
2994
02:25:46.460 --> 02:25:48.480
that really came out of nowhere in some ways,
2995
02:25:48.480 --> 02:25:52.050
and it was only possible because of this AI application
2996
02:25:52.050 --> 02:25:54.850
to identify mechanisms of actions.
2997
02:25:54.850 --> 02:25:56.400
So I'm gonna stop here.
2998
02:25:56.400 --> 02:25:58.090
And as I said, hopefully we can have a discussion
2999
02:25:58.090 --> 02:26:02.112
about how this could be used more broadly to analyze
3000
02:26:02.112 --> 02:26:04.980
other types of molecules and address
3001
02:26:04.980 --> 02:26:06.180
some addiction problems.
3002
02:26:08.700 --> 02:26:11.940
We have received several questions in the chat.
3003
02:26:11.940 --> 02:26:13.440
Some are already addressed, but it looks
3004
02:26:13.440 --> 02:26:15.360
like we'll have time to go through all of them.
3005
02:26:15.360 --> 02:26:16.310
So we can use this
3006
02:26:16.310 --> 02:26:18.760
as a chance to kind of expand on them and discuss.
3007
02:26:18.760 --> 02:26:21.053
I did get one question in a direct chat.
3008
02:26:21.972 --> 02:26:22.974
So I'll start with that
3009
02:26:22.974 --> 02:26:23.813
'cause I don't think anybody saw it.
3010
02:26:25.860 --> 02:26:30.140
So the question was for Dr. Elemento,
3011
02:26:30.140 --> 02:26:32.320
it's how does information about cell heterogeneity
3012
02:26:32.320 --> 02:26:33.960
and organization and complex tissues
3013
02:26:33.960 --> 02:26:37.370
like the CNS contribute to current target prediction models?
3014
02:26:37.370 --> 02:26:40.660
Do you see an opportunity with single cell (indistinct)
3015
02:26:40.660 --> 02:26:43.910
like brain or HCA to better understanding mechanisms
3016
02:26:43.910 --> 02:26:45.223
of action of drugs?
3017
02:26:46.720 --> 02:26:48.610
Yeah, that's a wonderful question.
3018
02:26:48.610 --> 02:26:50.253
Thanks so much for asking this.
3019
02:26:51.174 --> 02:26:52.430
And I think the approach we're taking is kind
3020
02:26:52.430 --> 02:26:54.360
of a reductionist approach in some ways
3021
02:26:54.360 --> 02:26:56.000
where we're using sort of bits
3022
02:26:56.000 --> 02:27:00.140
of data on sort of I guess a tractable sort
3023
02:27:00.140 --> 02:27:03.570
of models and tractable systems to be able
3024
02:27:03.570 --> 02:27:06.210
to gather information about mechanisms of actions.
3025
02:27:06.210 --> 02:27:10.020
I think as you saw in kind of the application of some
3026
02:27:10.020 --> 02:27:12.150
of the findings to for example the design
3027
02:27:12.150 --> 02:27:14.850
of clinical trials, I think once you understand
3028
02:27:14.850 --> 02:27:15.900
mechanisms of actions,
3029
02:27:15.900 --> 02:27:19.300
I think you can really sort of explore the biology
3030
02:27:19.300 --> 02:27:21.900
of the cell types of biology, of the disease
3031
02:27:21.900 --> 02:27:25.620
that you want to target in ways that you couldn't before.
3032
02:27:25.620 --> 02:27:28.630
So I'd say, I think the research that I described
3033
02:27:28.630 --> 02:27:31.290
is very complimentary to I think what you're trying
3034
02:27:31.290 --> 02:27:34.560
to get at, which is once we have very precise mechanisms
3035
02:27:34.560 --> 02:27:36.540
of actions for a particular molecule,
3036
02:27:36.540 --> 02:27:38.723
then we can really understand what are the cell types
3037
02:27:38.723 --> 02:27:40.680
that may be more likely to respond
3038
02:27:40.680 --> 02:27:43.940
or to sort of be affected by a particular treatment?
3039
02:27:43.940 --> 02:27:45.910
And that's I think very, very critical.
3040
02:27:45.910 --> 02:27:47.940
So I think very often we just don't have
3041
02:27:47.940 --> 02:27:50.480
this detailed mechanism of action that we need to have
3042
02:27:50.480 --> 02:27:51.420
to be able to understand
3043
02:27:51.420 --> 02:27:54.970
that some cell types will be effected while others may not,
3044
02:27:54.970 --> 02:27:56.080
and that could potentially give rise
3045
02:27:56.080 --> 02:27:59.663
to a toxicity as a result of unwanted targeting.
3046
02:28:01.750 --> 02:28:05.180
I could really echo that answer in the COVID-19 work,
3047
02:28:05.180 --> 02:28:08.060
by understanding the mechanisms of pathogenicity,
3048
02:28:08.060 --> 02:28:10.750
we identified well over a dozen existing drugs
3049
02:28:10.750 --> 02:28:13.670
that that could have an effect probably five
3050
02:28:13.670 --> 02:28:15.867
of which have already proven out
3051
02:28:15.867 --> 02:28:17.980
in clinical studies to be effective
3052
02:28:17.980 --> 02:28:20.400
in COVID-19 disease progression.
3053
02:28:20.400 --> 02:28:24.250
So that mechanistic molecular understanding
3054
02:28:24.250 --> 02:28:27.570
allows you to just really rocket
3055
02:28:27.570 --> 02:28:30.237
toward solutions in ways that we couldn't do before.
3056
02:28:34.058 --> 02:28:34.891
Another question we have
3057
02:28:34.891 --> 02:28:36.120
is for Dr. Jacobson.
3058
02:28:36.120 --> 02:28:38.540
Can you expand on how you're including co-morbidity
3059
02:28:38.540 --> 02:28:39.930
as an additional layer?
3060
02:28:39.930 --> 02:28:41.270
This could be particularly helpful
3061
02:28:41.270 --> 02:28:43.320
for psychiatric disorders like addiction.
3062
02:28:44.250 --> 02:28:45.830
Yeah, absolutely.
3063
02:28:45.830 --> 02:28:49.680
So we're using all these different layers
3064
02:28:49.680 --> 02:28:51.660
of clinical information
3065
02:28:51.660 --> 02:28:55.630
including other morbidities to predict a particular way
3066
02:28:55.630 --> 02:28:59.070
that I described for predicting gene expression
3067
02:28:59.070 --> 02:29:02.303
from all of the genes, we're using the matrix
3068
02:29:02.303 --> 02:29:06.620
in very large populations, clinical populations,
3069
02:29:06.620 --> 02:29:09.360
one example is in the VA where we have clinical records
3070
02:29:09.360 --> 02:29:11.810
for 23 million patients.
3071
02:29:11.810 --> 02:29:14.550
We're building these predictive, in this case,
3072
02:29:14.550 --> 02:29:19.320
morbidity networks, of what diseases are predictive
3073
02:29:19.320 --> 02:29:20.357
of other diseases.
3074
02:29:20.357 --> 02:29:23.180
And then you have the network topology
3075
02:29:23.180 --> 02:29:25.890
that allows you to extract out complex phenotypes.
3076
02:29:25.890 --> 02:29:27.900
And it's not just disease codes,
3077
02:29:27.900 --> 02:29:30.800
we're actually going through all
3078
02:29:30.800 --> 02:29:32.440
of the pharmacy-fill information,
3079
02:29:32.440 --> 02:29:34.960
all the lab information, everything
3080
02:29:34.960 --> 02:29:38.830
that's in the clinical record that can help us build up
3081
02:29:38.830 --> 02:29:41.500
hopefully even more clinically
3082
02:29:41.500 --> 02:29:43.340
and biologically-relevant phenotypes
3083
02:29:43.340 --> 02:29:44.700
than just the diagnostic codes,
3084
02:29:44.700 --> 02:29:47.910
which diagnostic codes are not always really reflective
3085
02:29:47.910 --> 02:29:49.670
of underlying biology,
3086
02:29:49.670 --> 02:29:52.300
they're sort of a clinical pigeonhole in many cases.
3087
02:29:52.300 --> 02:29:56.150
So we're trying to do data driven phenotype definition
3088
02:29:57.040 --> 02:30:00.070
and using AI and explainable AI to help build
3089
02:30:00.070 --> 02:30:02.950
more sophisticated phenotyping approaches
3090
02:30:02.950 --> 02:30:04.310
that then helps us
3091
02:30:04.310 --> 02:30:07.530
do better genotype-to-phenotype relationships.
3092
02:30:07.530 --> 02:30:10.360
But again, it's a holistic model that flows
3093
02:30:10.360 --> 02:30:11.530
in both directions.
3094
02:30:11.530 --> 02:30:13.380
Once we figure out a genetic architecture,
3095
02:30:13.380 --> 02:30:14.330
we can use that to see
3096
02:30:14.330 --> 02:30:17.150
if we can subtype different phenotypes as well.
3097
02:30:17.150 --> 02:30:19.470
So it's always sort of flowing back and forth
3098
02:30:19.470 --> 02:30:22.370
and getting more and more refined pictures of the biology.
3099
02:30:23.960 --> 02:30:26.000
The next question is also for you, Dr. Jacobson,
3100
02:30:26.000 --> 02:30:28.590
it's how much of the variance can be explained for ASD
3101
02:30:28.590 --> 02:30:29.840
with structural variance?
3102
02:30:33.560 --> 02:30:36.291
Referring a lot of that missing heritability,
3103
02:30:36.291 --> 02:30:37.124
I don't have the full capture,
3104
02:30:37.124 --> 02:30:39.980
but when we off the top of my head, but when we look
3105
02:30:39.980 --> 02:30:42.100
in those different categories of information,
3106
02:30:42.100 --> 02:30:44.810
we start to get really high penetrants.
3107
02:30:44.810 --> 02:30:48.069
You can start to actually get up in the numbers of 60,
3108
02:30:48.069 --> 02:30:53.069
70% penetrants mechanistically associating with ASD.
3109
02:30:54.330 --> 02:30:56.290
So that was really exciting to see
3110
02:30:56.290 --> 02:30:59.210
that that capturing the systems biology,
3111
02:30:59.210 --> 02:31:03.250
the heterogeneity with structural variance converges
3112
02:31:03.250 --> 02:31:06.415
towards being able to explain much more
3113
02:31:06.415 --> 02:31:08.340
of the story than we've been able to see with stamps.
3114
02:31:08.340 --> 02:31:10.300
So really high penetrants numbers.
3115
02:31:12.540 --> 02:31:15.080
Our next question is for Dr. Elemento.
3116
02:31:15.080 --> 02:31:16.430
In his analysis, has he been able
3117
02:31:16.430 --> 02:31:19.173
to identify sex-related differences in toxicity?
3118
02:31:20.320 --> 02:31:21.710
And that's another great question.
3119
02:31:21.710 --> 02:31:23.130
Thanks so much for asking this.
3120
02:31:23.130 --> 02:31:25.759
I think that's exactly the direction that we want
3121
02:31:25.759 --> 02:31:27.170
this research should take,
3122
02:31:27.170 --> 02:31:29.130
the idea of building these predictive models
3123
02:31:29.130 --> 02:31:31.860
using features for example, from the molecules,
3124
02:31:31.860 --> 02:31:34.940
but also from the host, from the targets
3125
02:31:34.940 --> 02:31:37.390
of the molecules to know where they are expressed
3126
02:31:38.257 --> 02:31:41.650
and so on, what tissues are expressed in and so on,
3127
02:31:41.650 --> 02:31:44.470
is really to be able to identify the features
3128
02:31:44.470 --> 02:31:46.820
that are important, that are predictive of toxicity
3129
02:31:46.820 --> 02:31:48.380
and then kinda go back
3130
02:31:48.380 --> 02:31:51.790
to different individuals and ask, what's the level
3131
02:31:51.790 --> 02:31:54.510
of variability among individuals of these features?
3132
02:31:54.510 --> 02:31:57.790
So if we see that a particular gene or a particular target
3133
02:31:57.790 --> 02:32:00.250
of a molecule is important in terms
3134
02:32:00.250 --> 02:32:03.430
of for example, predicting toxicity, do we see
3135
02:32:03.430 --> 02:32:06.510
that this target is more highly expressed in females
3136
02:32:06.510 --> 02:32:09.680
compared to males or less expressed?
3137
02:32:09.680 --> 02:32:10.550
I think that's exactly the kind
3138
02:32:10.550 --> 02:32:12.960
of mechanistic information that we can use
3139
02:32:12.960 --> 02:32:14.250
as part of the models.
3140
02:32:14.250 --> 02:32:16.210
It's this kind of model analysis if you want it,
3141
02:32:16.210 --> 02:32:18.980
which I think is sometimes I think undervalued when it comes
3142
02:32:18.980 --> 02:32:20.767
to these AI models.
3143
02:32:20.767 --> 02:32:23.060
The amount of information that we can get
3144
02:32:23.060 --> 02:32:24.940
from really understanding the features
3145
02:32:24.940 --> 02:32:27.550
and what they mean biologically
3146
02:32:27.550 --> 02:32:29.290
is as critical and as important
3147
02:32:29.290 --> 02:32:31.610
as building the models themselves
3148
02:32:31.610 --> 02:32:33.510
and showing that they are predictive.
3149
02:32:33.510 --> 02:32:35.370
So I think that's exactly the direction
3150
02:32:35.370 --> 02:32:37.290
that we wanna take is to use the models
3151
02:32:37.290 --> 02:32:40.440
as a way to down the line, identify
3152
02:32:40.440 --> 02:32:44.440
so groups of individuals who may be more likely
3153
02:32:44.440 --> 02:32:47.170
to experience toxicity and therefore in few week,
3154
02:32:47.170 --> 02:32:50.301
could potentially be sort of
3155
02:32:50.301 --> 02:32:54.390
stratified and in a way to maybe you avoid toxicity
3156
02:32:54.390 --> 02:32:56.963
you kind of maybe anticipated better than what we do now.
3157
02:33:03.460 --> 02:33:04.410
You had another question
3158
02:33:04.410 --> 02:33:06.530
for you that I don't believe was answered yet.
3159
02:33:06.530 --> 02:33:08.810
In place at predicting molecular targets,
3160
02:33:08.810 --> 02:33:10.530
Can ML algorithms be trained
3161
02:33:10.530 --> 02:33:13.370
on behavioral outcomes from human and animal studies?
3162
02:33:13.370 --> 02:33:15.940
Is there enough data out there for this?
3163
02:33:15.940 --> 02:33:19.663
I think this is a wonderful question as well.
3164
02:33:19.663 --> 02:33:21.490
I think as you saw my biases
3165
02:33:21.490 --> 02:33:24.800
towards oncology where what we're trying to do is to kill
3166
02:33:24.800 --> 02:33:26.100
the cancer cells.
3167
02:33:26.100 --> 02:33:27.160
But I think the reality
3168
02:33:27.160 --> 02:33:30.708
is that there's a tremendous amount of information
3169
02:33:30.708 --> 02:33:32.570
for example about the target of the molecule
3170
02:33:32.570 --> 02:33:37.081
and in in sort of the behavioral sort of profile,
3171
02:33:37.081 --> 02:33:38.990
the modification of behaviors profile.
3172
02:33:38.990 --> 02:33:41.070
And if you have that information for many, many molecules
3173
02:33:41.070 --> 02:33:42.110
I think, just like I said,
3174
02:33:42.110 --> 02:33:44.190
you can position the new molecules
3175
02:33:44.190 --> 02:33:47.380
in the kind of universe of these, you have all sort
3176
02:33:47.380 --> 02:33:49.780
of changes, modifications, and therefore
3177
02:33:49.780 --> 02:33:51.560
I think you should be able to use that information
3178
02:33:51.560 --> 02:33:55.090
as a way to improve how you sort
3179
02:33:55.090 --> 02:33:56.950
of predict picking in terms of actions,
3180
02:33:56.950 --> 02:33:59.410
especially for, again for molecules that don't kill cells,
3181
02:33:59.410 --> 02:34:03.790
that have an activity that is more focused
3182
02:34:03.790 --> 02:34:07.033
on for example, the addiction biology.
3183
02:34:08.974 --> 02:34:10.910
The next question is directed at both Dr. Xu
3184
02:34:10.910 --> 02:34:11.747
and Dr. Narain.
3185
02:34:11.747 --> 02:34:14.360
Do you also include animal behavioral studies
3186
02:34:14.360 --> 02:34:17.360
in the literature for the testing stages in your algorithms?
3187
02:34:24.100 --> 02:34:29.100
Okay, so we do include a lot of animal phenotype data
3188
02:34:34.710 --> 02:34:38.230
like from (indistinct) experiment.
3189
02:34:38.230 --> 02:34:43.230
So, but is not from the literature, is from the database.
3190
02:34:50.130 --> 02:34:55.130
So in the (indistinct) the systematic (indistinct)
3191
02:34:55.320 --> 02:34:57.280
was held in the gene in the mouse model.
3192
02:34:57.280 --> 02:34:59.860
And then they observed a lot of phenotype.
3193
02:34:59.860 --> 02:35:02.780
So many of the phenotype are related
3194
02:35:02.780 --> 02:35:07.720
to like behavior and also even like addiction.
3195
02:35:07.720 --> 02:35:11.510
So we do use a lot of those mouse phenotype
3196
02:35:11.510 --> 02:35:15.460
for perform like a virtual from a typical screen
3197
02:35:15.460 --> 02:35:17.603
or for targeted drugs.
3198
02:35:20.460 --> 02:35:23.060
And for us what we've done,
3199
02:35:23.060 --> 02:35:26.090
so with respect to the model bills
3200
02:35:26.090 --> 02:35:28.690
and the algorithms, everything is human-based.
3201
02:35:28.690 --> 02:35:31.450
We use the animal studies or animal trials
3202
02:35:31.450 --> 02:35:34.690
or the behavioral mechanisms at some of the laboratories,
3203
02:35:34.690 --> 02:35:37.130
like the Nestler Laboratory at Mount Sinai,
3204
02:35:37.130 --> 02:35:39.473
or psychogenics as a validation point.
3205
02:35:41.670 --> 02:35:42.503
Okay.
3206
02:35:42.503 --> 02:35:45.000
The next question is for Dr. Elemento.
3207
02:35:45.000 --> 02:35:47.830
Based on your finding of D2R target, have you tested
3208
02:35:47.830 --> 02:35:51.110
the potential value of other D2R agonists and glioma
3209
02:35:51.110 --> 02:35:54.227
that express D2R other cancers?
3210
02:35:54.227 --> 02:35:57.754
Antagonist, antagonist, not agonist.
3211
02:35:57.754 --> 02:35:58.587
Awesome.
3212
02:35:58.587 --> 02:35:59.710
I can just, yeah.
3213
02:35:59.710 --> 02:36:01.620
Yeah, that's wonderful question too.
3214
02:36:01.620 --> 02:36:03.110
Lots of (laughing) great question today.
3215
02:36:03.110 --> 02:36:04.210
That's awesome.
3216
02:36:04.210 --> 02:36:07.370
And we actually have not and partially,
3217
02:36:07.370 --> 02:36:11.140
well, actually, so part of what the experiments came out
3218
02:36:12.200 --> 02:36:13.870
out of the validation that I didn't really have
3219
02:36:13.870 --> 02:36:15.550
a chance to kind of explain to details
3220
02:36:15.550 --> 02:36:19.380
is that what makes this molecule ONC201 special
3221
02:36:19.380 --> 02:36:22.760
is actually not only the fact that it targets DRD2,
3222
02:36:22.760 --> 02:36:25.650
but the fact that it doesn't really target DRD5
3223
02:36:25.650 --> 02:36:27.818
which is a kind of homolog
3224
02:36:27.818 --> 02:36:29.880
of DRD2, which I think most
3225
02:36:29.880 --> 02:36:33.250
of the dopamine receptors antagonists actually target both.
3226
02:36:33.250 --> 02:36:36.680
So we think that the reason why this molecule is special
3227
02:36:36.680 --> 02:36:38.790
and has maybe these anticancer activity
3228
02:36:38.790 --> 02:36:40.300
that has not really been observed
3229
02:36:40.300 --> 02:36:43.850
with the other dopamine receptor antagonist
3230
02:36:43.850 --> 02:36:45.860
is really this kind of differential activity
3231
02:36:45.860 --> 02:36:50.220
and the two dopamine receptor molecules or proteins.
3232
02:36:50.220 --> 02:36:54.674
And that's why we haven't sort of tested ourselves
3233
02:36:54.674 --> 02:36:58.780
these other molecules in gliomas,
3234
02:36:58.780 --> 02:37:01.530
really mostly because we think that ONC201 has a kind
3235
02:37:01.530 --> 02:37:03.330
of very special mechanism of action.
3236
02:37:05.410 --> 02:37:09.120
But it will be interesting to kind of (indistinct)
3237
02:37:12.165 --> 02:37:13.580
I think we've answered all the questions
3238
02:37:13.580 --> 02:37:15.900
that hadn't been answered, but we still have a few minutes.
3239
02:37:15.900 --> 02:37:18.220
So I guess I'll circle back to the earlier questions
3240
02:37:18.220 --> 02:37:19.160
that were answered in the chat
3241
02:37:19.160 --> 02:37:21.270
just in case anyone wants to expand on them.
3242
02:37:21.270 --> 02:37:24.300
So there was a question to Dr. Xu, it seems the approach
3243
02:37:24.300 --> 02:37:26.670
taken by AI might just find too many drugs.
3244
02:37:26.670 --> 02:37:27.973
How can this be avoided?
3245
02:37:29.520 --> 02:37:32.502
Yeah, that's a great question.
3246
02:37:32.502 --> 02:37:37.370
So basically so depending on your goal
3247
02:37:37.370 --> 02:37:41.333
you want from the me-too drugs for you,
3248
02:37:43.025 --> 02:37:45.470
or you want to find something totally new,
3249
02:37:45.470 --> 02:37:50.320
I expect, so you can build a different algorithm for each
3250
02:37:50.320 --> 02:37:54.033
of these, depends on the purpose.
3251
02:37:54.870 --> 02:37:58.830
The problem with things funded totally a lot of things
3252
02:37:59.720 --> 02:38:01.603
is work hard to evaluate.
3253
02:38:02.571 --> 02:38:05.930
So you say, or you can go to the ELMo model
3254
02:38:05.930 --> 02:38:09.970
to test the those, but that's can be very expensive.
3255
02:38:09.970 --> 02:38:11.883
And also ELMo model, as I said,
3256
02:38:13.049 --> 02:38:14.030
there's a bigger limitation
3257
02:38:14.030 --> 02:38:16.880
for drug discovery, because especially
3258
02:38:16.880 --> 02:38:18.430
for neuro-psychiatric disorders,
3259
02:38:18.430 --> 02:38:22.000
because the ELMo model is way different from human diseases.
3260
02:38:22.000 --> 02:38:27.000
So it's not not the only way to evaluate a drug context.
3261
02:38:31.110 --> 02:38:31.943
Thank you.
3262
02:38:31.943 --> 02:38:33.180
I'm gonna ask if Nora and George
3263
02:38:33.180 --> 02:38:35.160
have any further questions since we're getting close
3264
02:38:35.160 --> 02:38:36.053
on the time.
3265
02:38:39.120 --> 02:38:41.553
I asked my questions, thanks.
3266
02:38:46.130 --> 02:38:48.183
Nope, no more questions from me.
3267
02:38:50.817 --> 02:38:51.830
Thank you everyone.
3268
02:38:51.830 --> 02:38:52.663
I wanna thank our speakers again,
3269
02:38:52.663 --> 02:38:54.050
and thank my co-chair Vani
3270
02:38:54.050 --> 02:38:55.070
And also thank the audience
3271
02:38:55.070 --> 02:38:56.140
for all of these great questions.
3272
02:38:56.140 --> 02:38:58.540
I'm gonna pass it back to John.
3273
02:38:58.540 --> 02:39:01.400
Okay, thank you, another great session.
3274
02:39:01.400 --> 02:39:04.540
It's my privilege to introduce again
3275
02:39:04.540 --> 02:39:08.770
the Institute Directors of NIDA and NIAAA,
3276
02:39:08.770 --> 02:39:12.673
Nora Volkow and George Koob to make a few comments.
3277
02:39:16.800 --> 02:39:19.070
George, do you want to go first?
3278
02:39:19.070 --> 02:39:22.393
Yeah, so let me share my screen.
3279
02:39:24.050 --> 02:39:27.683
I wanna start by thanking everyone.
3280
02:39:36.931 --> 02:39:38.381
If I can get this to show up.
3281
02:39:45.070 --> 02:39:46.263
Is it showing up?
3282
02:39:49.770 --> 02:39:51.020
Anybody see?
3283
02:39:51.020 --> 02:39:52.270
Not all the way.
3284
02:39:54.520 --> 02:39:55.570
Okay, hang on.
3285
02:39:55.570 --> 02:39:57.663
We just got a little sliver there.
3286
02:40:00.960 --> 02:40:01.913
Are we better?
3287
02:40:02.930 --> 02:40:03.763
Yup.
3288
02:40:04.960 --> 02:40:06.920
Go to slide presentation.
3289
02:40:06.920 --> 02:40:08.533
Yeah, that's what I gotta do.
3290
02:40:11.771 --> 02:40:14.230
You think by now I'd get this down.
3291
02:40:14.230 --> 02:40:15.510
All right.
3292
02:40:15.510 --> 02:40:19.093
So thanks to Roger, and I wanna just thank everybody
3293
02:40:19.093 --> 02:40:23.630
at NIDA and NIAAA and the Neuroscience Work Group.
3294
02:40:23.630 --> 02:40:27.060
The mission of the NIDA-NIAAA Neuroscience Work Group
3295
02:40:27.060 --> 02:40:29.260
is to provide a forum to facilitate
3296
02:40:29.260 --> 02:40:30.460
the discussion development
3297
02:40:30.460 --> 02:40:32.710
of neuroscience research programs, to understand,
3298
02:40:32.710 --> 02:40:36.300
prevent and treat substance abuse and addiction.
3299
02:40:36.300 --> 02:40:38.980
The work group consists of extramural and intramural staff
3300
02:40:38.980 --> 02:40:41.870
from NIDA and NIAAA who have brought interest
3301
02:40:41.870 --> 02:40:44.450
and understanding the neuroscience of addiction.
3302
02:40:44.450 --> 02:40:46.551
And I just want you to see the extent
3303
02:40:46.551 --> 02:40:49.610
of the commitment from both institutes on this slide,
3304
02:40:49.610 --> 02:40:51.580
and it's extraordinary.
3305
02:40:51.580 --> 02:40:53.180
And it makes me really proud
3306
02:40:53.180 --> 02:40:57.150
of our collaborative research on the addictions at NIH
3307
02:40:57.150 --> 02:41:00.090
which we started some seven years ago,
3308
02:41:00.090 --> 02:41:01.590
believe it or not Nora.
3309
02:41:01.590 --> 02:41:04.570
And I wanna make a real shout-out also
3310
02:41:04.570 --> 02:41:08.850
to the stars behind the scenes and everyone
3311
02:41:08.850 --> 02:41:13.850
who's been involved (indistinct) Susan Holbrook, David Mazza
3312
02:41:13.890 --> 02:41:15.453
and Caitlin Dudevoir.
3313
02:41:17.579 --> 02:41:18.490
Dudevoir
3314
02:41:18.490 --> 02:41:21.000
I should be able to pronounce that since I speak French.
3315
02:41:21.000 --> 02:41:24.520
Anyway, the only other thing I wanted to say
3316
02:41:24.520 --> 02:41:27.480
other than thanking everyone is I've just been blown
3317
02:41:27.480 --> 02:41:31.820
away by today's and yesterday's presentations.
3318
02:41:31.820 --> 02:41:34.620
I really enjoyed the last one.
3319
02:41:34.620 --> 02:41:36.031
It just opens up,
3320
02:41:36.031 --> 02:41:41.031
and the young people's Career Investigators Showcase,
3321
02:41:41.040 --> 02:41:43.790
'cause it just opened up so many new avenues for us.
3322
02:41:43.790 --> 02:41:46.010
And what I took home
3323
02:41:46.010 --> 02:41:49.750
from this frontiers that we haven't done before,
3324
02:41:49.750 --> 02:41:53.940
it's we were talking about relationships
3325
02:41:53.940 --> 02:41:57.630
at many, many different levels, hierarchical
3326
02:41:57.630 --> 02:41:59.870
but also sideways, mining
3327
02:41:59.870 --> 02:42:04.870
those relationships back-translation, pharmacoepidemiology.
3328
02:42:05.030 --> 02:42:06.100
Who would have ever dreamed
3329
02:42:06.100 --> 02:42:08.250
we'd be talking about pharmacoepidemiology?
3330
02:42:09.800 --> 02:42:13.370
Revisiting our metrics, biomarkers,
3331
02:42:13.370 --> 02:42:18.370
and all of it today and yesterday reflects
3332
02:42:19.280 --> 02:42:23.061
on our mission or at least the way I look at our mission
3333
02:42:23.061 --> 02:42:25.323
which is the vulnerability, the diagnosis and the treatment
3334
02:42:25.323 --> 02:42:29.440
and the prevention of substance use disorders
3335
02:42:29.440 --> 02:42:31.000
and of course our institute
3336
02:42:31.913 --> 02:42:32.746
with the focus on alcohol use disorder.
3337
02:42:32.746 --> 02:42:34.120
So I'll stop there.
3338
02:42:34.120 --> 02:42:36.640
I've just really enjoyed this very much.
3339
02:42:36.640 --> 02:42:38.250
I think this has to go down as one
3340
02:42:38.250 --> 02:42:43.250
of my favorite Frontier's programs that we put together.
3341
02:42:43.785 --> 02:42:46.820
And I wanna thank everybody again for your excellent talk.
3342
02:42:46.820 --> 02:42:50.110
And some of you, I didn't get to thank personally
3343
02:42:50.110 --> 02:42:53.960
through the chat, but I don't think there was one bad talk
3344
02:42:53.960 --> 02:42:56.590
and they were all outstanding from my perspective.
3345
02:42:56.590 --> 02:42:59.120
So I'll turn it over to you, Nora.
3346
02:42:59.120 --> 02:43:00.610
Yeah, no, and I completely agree.
3347
02:43:00.610 --> 02:43:04.200
And I also certainly want thank the enormous amount
3348
02:43:04.200 --> 02:43:05.130
of effort that went
3349
02:43:05.130 --> 02:43:08.890
into these extraordinary successful event.
3350
02:43:08.890 --> 02:43:12.483
And it's sort of to me, is sort of like the sense of high,
3351
02:43:12.483 --> 02:43:13.870
and the sense of high
3352
02:43:13.870 --> 02:43:17.990
because on Wednesday we all leaved a very tragic event.
3353
02:43:17.990 --> 02:43:21.640
And then Thursday and Friday, we've been confronted
3354
02:43:21.640 --> 02:43:24.720
with the other side of human creativity
3355
02:43:24.720 --> 02:43:28.480
and ability to transform things by using science
3356
02:43:28.480 --> 02:43:32.340
and actually importantly collaborations and partnerships.
3357
02:43:32.340 --> 02:43:34.860
So the beauty
3358
02:43:34.860 --> 02:43:37.140
of the science that was presented today
3359
02:43:37.140 --> 02:43:39.937
and yesterday speaks for itself.
3360
02:43:39.937 --> 02:43:42.690
And it crystallizes the notion that we now
3361
02:43:42.690 --> 02:43:47.360
have the tools clearly to start to look into complexity.
3362
02:43:47.360 --> 02:43:50.490
And as we go more into these complex tools,
3363
02:43:50.490 --> 02:43:54.565
these will proliferate on what we'll be able to do two
3364
02:43:54.565 --> 02:43:55.700
or three years from now
3365
02:43:55.700 --> 02:43:57.637
are things that we cannot predict right now.
3366
02:43:57.637 --> 02:44:01.140
And it's going to change our ability to bring new treatments
3367
02:44:01.140 --> 02:44:03.840
and bring our ability to follow people
3368
02:44:03.840 --> 02:44:05.737
to do a proper personalized interventions.
3369
02:44:05.737 --> 02:44:09.460
And so eloquently was presented and importantly
3370
02:44:09.460 --> 02:44:13.516
our ability to understand that much greater that a disease
3371
02:44:13.516 --> 02:44:15.440
and the interactions of genes and environments
3372
02:44:15.440 --> 02:44:18.790
and our unique trajectories for each one of us.
3373
02:44:18.790 --> 02:44:21.650
So how can you not be excited about it?
3374
02:44:21.650 --> 02:44:22.840
So I am very excited.
3375
02:44:22.840 --> 02:44:25.550
So thanks all of the presenters for that.
3376
02:44:25.550 --> 02:44:28.960
But the other reason why I'm so happy at these
3377
02:44:28.960 --> 02:44:33.030
is that I've seen now how well both of the institutes
3378
02:44:33.030 --> 02:44:36.200
are working together, which is the way that it should be.
3379
02:44:36.200 --> 02:44:38.130
Science is the product
3380
02:44:38.130 --> 02:44:40.000
of collaborations, of partnerships,
3381
02:44:40.000 --> 02:44:41.960
so people thinking together,
3382
02:44:41.960 --> 02:44:43.960
and it make some enormous amounts
3383
02:44:43.960 --> 02:44:47.160
of sense for all of us to work together
3384
02:44:47.160 --> 02:44:50.410
from the NIH, from the institutes, from the investigators
3385
02:44:50.410 --> 02:44:55.240
because we all have a common goal and the uniqueness
3386
02:44:55.240 --> 02:44:57.860
that each individual brings actually adds
3387
02:44:57.860 --> 02:45:00.460
to the value of where we can go.
3388
02:45:00.460 --> 02:45:05.460
So I'm delighted with the events today and yesterday.
3389
02:45:05.460 --> 02:45:07.800
It's very, very exciting to see it,
3390
02:45:07.800 --> 02:45:12.800
but I do want thank again, my colleague,
3391
02:45:13.427 --> 02:45:16.750
Dr. George Cobb which actually has embraced
3392
02:45:16.750 --> 02:45:20.650
also these the importance of the working together
3393
02:45:20.650 --> 02:45:23.460
of the two institutes and has been able
3394
02:45:23.460 --> 02:45:26.880
through leadership also to generate these,
3395
02:45:26.880 --> 02:45:30.000
but certainly also within each institute,
3396
02:45:30.000 --> 02:45:33.030
I'm also grateful for what you have done.
3397
02:45:33.030 --> 02:45:37.540
So without having said, I thank you all
3398
02:45:37.540 --> 02:45:41.337
for participating and for the work that went into it.
3399
02:45:41.337 --> 02:45:43.060
And I look forward
3400
02:45:43.060 --> 02:45:44.860
to the new findings that are going to start
3401
02:45:44.860 --> 02:45:46.970
to that are already emerging
3402
02:45:46.970 --> 02:45:51.590
but that are gong to be just blooming up our understanding
3403
02:45:51.590 --> 02:45:53.123
of substance use disorders.
3404
02:45:54.460 --> 02:45:56.310
So thanks-
Nora, I just wanted to say,
3405
02:45:56.310 --> 02:45:58.193
I couldn't have done it without you.
3406
02:45:59.717 --> 02:46:02.950
And I really wanna put a shout-out to Roger Sorenson
3407
02:46:02.950 --> 02:46:04.740
and John Matochik.
3408
02:46:04.740 --> 02:46:07.950
They've just put an enormous amount of energy into this
3409
02:46:07.950 --> 02:46:10.210
as well as all the colleagues we've talked about.
3410
02:46:10.210 --> 02:46:12.050
So Bravo everybody (clapping)
3411
02:46:12.050 --> 02:46:17.050
And think about this over the weekend as Nora said,
3412
02:46:17.250 --> 02:46:18.900
rather than all the other things (laughing)
3413
02:46:18.900 --> 02:46:20.893
that we've had to deal with this week.
3414
02:46:24.770 --> 02:46:26.920
So bye bye (laughing)
3415
02:46:26.920 --> 02:46:27.880
Stay safe, sign off (laughing)
3416
02:46:27.880 --> 02:46:31.830
See you everybody at the next mini-convention, take care.
3417
02:46:31.830 --> 02:46:34.670
I'll try, we'll see you in person hopefully next time.
3418
02:46:34.670 --> 02:46:35.608
Thanks.
3419
02:46:35.608 --> 02:46:36.780
You will see us in person (laughing)
3420
02:46:36.780 --> 02:46:39.836
Yeah, it would be nice to see you in person.
3421
02:46:39.836 --> 02:46:42.879
(Roger laughing)
3422
02:46:42.879 --> 02:46:43.716
Thanks All.