You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

3426 lines
120 KiB

WEBVTT
01:08.212 --> 01:09.332
Let's try this again.
01:09.352 --> 01:16.193
I know a lot of people that watch are disappointed that I can't keep my cool.
01:17.514 --> 01:21.575
I know that a lot of people are disappointed that I can't remain calm.
01:23.717 --> 01:36.890
But when you get on line with somebody and you know and you start to feel that they're being disingenuous that they're really just lying about what their motives are, what they wanted to accomplish and what they really meant to do.
01:39.212 --> 01:42.733
It starts to become very obvious if people have tried this with you before.
01:43.474 --> 01:46.494
And you know what the trick that they were trying to pull today, I think, is?
01:46.675 --> 01:49.735
And I'm just gonna say it out loud because the truth hurts them.
01:50.636 --> 01:54.577
They thought maybe that a guy with an accent could be used to do it.
01:55.797 --> 02:05.600
Because anybody that can hide behind, oh, I didn't understand, or I didn't quite get the English right, or I didn't realize you were attacking all anonymous accounts.
02:05.640 --> 02:06.581
You were attacking me!
02:07.481 --> 02:11.065
And because I supported you before, then you knew who I was.
02:11.505 --> 02:12.646
This is all a story.
02:12.706 --> 02:14.748
This has no basis in reality.
02:15.569 --> 02:18.452
And these people are anonymously online.
02:18.652 --> 02:24.959
So many foreigners are anonymously online competing for the consciousness of Americans.
02:26.021 --> 02:27.362
What's he doing in France?
02:28.262 --> 02:32.505
What did he do in France to speak out against what's gone on there?
02:32.565 --> 02:33.706
What is he doing in France?
02:33.766 --> 02:37.788
Why is he so interested in putting a crack pipe in an American's mouth?
02:37.848 --> 02:44.272
Why isn't he tweeting in his natural language?
02:45.673 --> 02:49.455
It becomes pretty interesting that he didn't want to talk about any of these questions.
02:49.515 --> 02:52.857
He wanted to talk about whether Nick got Pfizer right or not.
02:54.172 --> 03:00.799
It's interesting that he didn't want to talk about any of the biology that we've been fooled into believing.
03:01.820 --> 03:04.723
He didn't want to talk about murder over and over again.
03:04.763 --> 03:08.466
He avoided talking about, oh, we can talk about when they did the murder, he said.
03:10.528 --> 03:14.532
And he tried so hard to downplay the crack pipe.
03:15.407 --> 03:17.108
But you have to see it for what it is.
03:17.809 --> 03:21.772
My tweet to him was a statement.
03:22.593 --> 03:30.038
His tweet to me when I said that they don't call it murder was exactly the opposite.
03:31.579 --> 03:34.121
And I probably made a mistake by having the guy up.
03:34.362 --> 03:38.064
I mean, you don't want to go into a sword fight with somebody who's going to bring a gun.
03:39.245 --> 03:42.128
You don't want to go into an argument that somebody who's going to lie
03:44.639 --> 03:47.281
And that's why I gave a long introduction before he came in.
03:47.321 --> 03:51.864
That's why I'm repeating it because these people don't want you to think about the history.
03:51.924 --> 03:57.468
They don't want you to think, talk about, about Kaprowski and Plotkin and Gallo and Baltimore.
03:57.988 --> 04:04.392
They don't want us to talk about how, how Kevin McKernan's dad was probably having barbecues with those people.
04:04.913 --> 04:09.276
They don't want us to talk about how Sasha Latipova is a bad guy.
04:09.456 --> 04:10.216
Oh, maybe they do.
04:11.097 --> 04:12.358
Maybe that's why he brought it up.
04:13.120 --> 04:20.665
because that's also part of this little trap that they want us to get into so that we can't actually escape the mythology.
04:22.327 --> 04:25.189
Notice that the guy never came with anything concrete.
04:25.609 --> 04:27.210
He didn't want to talk about the biology.
04:27.250 --> 04:29.292
He wanted to talk about why I attacked him.
04:30.793 --> 04:41.500
And once you see that everything on the internet is aimed at creating an illusion, an artificial wave that's really fun to surf,
04:42.478 --> 04:47.440
An artificial wave of ideas that seems to be spontaneously happening.
04:48.400 --> 04:54.063
But in reality, they know exactly how it's happening and the people who are doing it know exactly what they're trying to do.
04:54.603 --> 04:59.765
They're trying to sustain this spectacular commitment to lies.
05:00.645 --> 05:02.546
We were gonna talk about Pfizer.
05:03.266 --> 05:04.687
You gotta be kidding me.
05:05.147 --> 05:06.588
You gotta be kidding me.
05:08.959 --> 05:10.040
But I mean, I should have known.
05:10.060 --> 05:12.001
I should have known that that was what it was going to be.
05:12.141 --> 05:13.321
I should have known.
05:15.162 --> 05:15.843
I should have known.
05:16.443 --> 05:17.744
Stay focused on the biology.
05:17.784 --> 05:20.225
Don't take the bait on social media, which I just did.
05:20.866 --> 05:22.186
And love your neighbor.
05:23.747 --> 05:26.268
I'm not going to defend anonymity on the internet.
05:26.288 --> 05:30.131
You can be anonymous on the internet and at the same time have a public-facing account.
05:30.171 --> 05:30.731
I don't care.
05:31.171 --> 05:35.814
But if you don't have the guts to stand up, he said that standing up wouldn't do any good for him.
05:36.894 --> 05:39.816
That's just the kind of mealy-mouthed excuse that got us here.
05:41.017 --> 05:42.898
Wouldn't do me any good to stand up.
05:42.978 --> 05:48.422
I've got expertise, but I don't really like to... I don't really like to lord it over anybody.
05:50.063 --> 05:52.605
About the fakest anybody could have been.
05:53.126 --> 05:54.927
They were counting on me getting fired up.
05:55.747 --> 05:59.930
They were counting on me getting fired up because they know that that's how I am.
05:59.990 --> 06:01.972
That's my weakness, I guess.
06:03.717 --> 06:06.338
I don't think it's a weakness because it's still telling the truth.
06:06.598 --> 06:08.618
I get upset because these people are liars.
06:08.718 --> 06:10.199
I get upset because I can hear it.
06:10.739 --> 06:11.559
I can sense it.
06:12.059 --> 06:15.100
And I can see what they're trying to do with this limited spectrum of debate.
06:16.080 --> 06:20.101
I could have sent you talking points, he said, and then you would know what we're going to talk about.
06:20.181 --> 06:20.742
No, no, no.
06:20.842 --> 06:24.042
I'm not complaining that I didn't know what we were going to talk about.
06:24.082 --> 06:25.983
I'm complaining what you chose to talk about.
06:27.711 --> 06:31.800
Just pathetic illusion created by anonymous accounts on the internet.
06:31.860 --> 06:33.143
Probably he has several.
06:33.504 --> 06:36.029
Maybe he's five different mouse accounts for all we know.
06:36.651 --> 06:37.452
That's how they do it.
06:39.923 --> 06:40.103
do
07:05.040 --> 07:10.645
It is take two for the 6th of June, 2024.
07:10.965 --> 07:12.527
Take two, it's 1253.
07:12.587 --> 07:15.049
We've got about an hour, an hour and a half.
07:15.889 --> 07:18.151
I would like to do a little bit of concrete work here.
07:18.652 --> 07:19.753
I'm a human just like you.
07:19.913 --> 07:26.559
I am vulnerable to the internet and to some of this harassment if I pay enough attention to it, and that's what happened earlier.
07:28.380 --> 07:29.942
I'm going to be at the Red Pill Expo.
07:29.982 --> 07:35.348
There is a virtual Red Pill Expo that you can sign up for, and I hope that you all will do that.
07:35.428 --> 07:40.333
Remember, intramuscular injection of any combination of substances with the intent of augmenting the immune system is dumb.
07:40.953 --> 07:42.956
Cancelled mouse didn't want to talk about that.
07:43.656 --> 07:48.620
Transfection in healthy humans was always criminally negligent and Cancelled Mouse didn't want to talk about that either.
07:49.261 --> 07:53.204
RNA cannot pandemic and Cancelled Mouse didn't want to talk about that either.
07:53.244 --> 08:08.056
He wanted to address why it was that I attacked him as opposed to all the other anonymous accounts that I was actually attacking and suggesting that you should try not to pay attention to anonymous accounts because it's turning your consciousness over to the slavers.
08:09.198 --> 08:12.381
And he tried to twist that around into being an attack on him.
08:13.122 --> 08:14.844
And then he tried to make it about Nick.
08:15.104 --> 08:16.906
And then he tried to make it about Pfizer.
08:16.946 --> 08:18.788
And then he tried to make it about Sasha.
08:26.577 --> 08:27.778
I'm sure we're under attack.
08:30.598 --> 08:38.962
And I'm sure that we are under attack by a lot of different angles, but one of them is anonymous accounts that pretend to be good.
08:39.062 --> 08:45.526
Anonymous accounts that are responsible for the spread of very bad ideas for the last three years.
08:45.566 --> 08:51.829
That's why we heard Kevin McKernan say repeatedly in 2020 that where he's doing all his debating is on Twitter.
08:52.189 --> 08:57.532
That's why we've heard throughout the pandemic that Twitter is the town square, that that's where free speech is being,
08:58.272 --> 09:00.174
is being preserved by Elon Musk.
09:00.214 --> 09:04.239
And it's these anonymous accounts, apparently, that are part of this free speech.
09:04.680 --> 09:10.867
It couldn't possibly be foreign meddlers trying to destroy or assist in the controlled demolition of America.
09:10.947 --> 09:11.968
That would be crazy.
09:14.597 --> 09:19.100
So anyway, I'm going to keep making these points because apparently that's what they don't want me to do.
09:19.260 --> 09:23.703
So I think that these people are working against us with a coordinated symphony of lying.
09:23.743 --> 09:34.151
It's a little bit of like a Lollapalooza of liars and it doesn't matter what stage you go to, you're always going to hear a limited spectrum of debate where they argue very vigorously and occasionally drop the names of some of the people that you
09:34.671 --> 09:57.669
you expect them to be listening to but in reality they are carefully curating a narrative about a gain-of-function virus that will likely come again so that our children believe that pandemic potential is real and this limited spectrum of debate is indeed curated on the internet on social media now in a way that it could never have been curated before there are videos from as late as two thousand
09:59.486 --> 10:27.781
19 where they're lamenting about how the swine flu could have gone a lot better in the and the Conformity that we could have gotten would have been a lot better if we had Facebook or anything like we have now And this organized and intelligent Manipulation of the conscious habits and opinions of the masses is at a new level It's an absolutely a new level and
10:29.403 --> 10:31.144
And it is being done by these people.
10:31.184 --> 10:34.025
And again, remember, Canceled Mouse didn't want to address any of this.
10:34.065 --> 10:35.926
He didn't want to talk about the bad guys.
10:35.966 --> 10:49.752
He didn't want to talk about the bad guy, Kevin McKernan, that I've done three days, five days in a row on now about all these different crazy podcasts that he was on, where he was extraordinarily similar to a pot
10:50.752 --> 11:15.737
Associated Brett Weinstein in his messaging about masks and about PCR And never anywhere in there is this idea that maybe the national security state is lying about it Oh, no, wait one time they said it, but he doesn't think so and so, of course this guy was promoted repeatedly by CHD CHD did two different videos with him one with Mary Holland and and
11:17.962 --> 11:23.147
and Brian Hooker, where they talked about the DNA double, the DNA contamination of the shots.
11:25.549 --> 11:32.335
And we have, of course, these people that I've been calling out, Canceled Mouse didn't want to talk about Jessica Rose or Robert Malone.
11:33.295 --> 11:44.025
He just vaguely said that we agree about most stuff, but since we agree about most stuff, it doesn't really make sense why somebody who agrees with about most stuff would put a crack pipe in my mouth on Twitter.
11:44.485 --> 11:47.246
because I attacked anonymous accounts.
11:47.706 --> 11:51.647
And I'm one of those anonymous accounts and you actually quoted my anonymous account.
11:52.007 --> 11:56.129
So therefore you must have been attacking me because I used to support you or something like that.
11:57.369 --> 12:12.374
And this is all part of the same illusion, whether he knows it or not, he's contributing to this chaos on the internet that is misleading the young about the potential for pandemics that can be contained in a particular RNA virus.
12:14.366 --> 12:18.431
And I have a personal vendetta with Brett Weinstein.
12:18.451 --> 12:21.255
I can't wait to shake his hand in November when he's here in Pittsburgh.
12:22.376 --> 12:27.523
And I have kind of a personal vendetta with anybody that attacks me while promoting him.
12:28.544 --> 12:30.546
And I have a suspicion that that particular
12:33.513 --> 12:38.699
Polish man up there who used to work for the CDC is part of this controlled operation.
12:38.779 --> 12:43.925
And there are other people, I think, that are close to me that are probably involved in this as well.
12:43.965 --> 12:47.949
And they're just better at not being exposed or better at staying hidden.
12:48.830 --> 12:53.675
And so it could be that Cancelled Mouse is right and that maybe Nick Hudson is also playing for somebody else.
12:53.715 --> 12:54.476
But Nick Hudson,
12:55.461 --> 13:02.447
has promoted my interview with Wolfgang Wodach as one of the best interviews during the pandemic, the best.
13:02.587 --> 13:05.049
And there is a sub stack for it if you're interested.
13:06.030 --> 13:22.505
And so unlike canceled mouse who puts a crack pipe in my mouth when I say that the mice won't say it's murder, and then tries to go back to another tweet from many, many, many weeks ago saying that you attacked me here when I actually said that anonymous accounts were a bad idea and following them was horrible.
13:23.697 --> 13:25.937
And I called Twitter a cesspool of an app.
13:27.518 --> 13:28.478
And so what did he do?
13:28.538 --> 13:30.558
He really did a little bait and switch there.
13:30.738 --> 13:37.040
He put a crack pipe in my mouth because he wanted to attribute bad ideas to me.
13:37.980 --> 13:44.801
And the tweet that I did under there, which of course he says is significant because I quoted his tweet, that meant I was attacking him.
13:45.361 --> 13:48.062
But when he quotes my tweet, it's just a general statement.
13:49.209 --> 13:54.651
It's just a general statement because you attacked me, but I attacked you like four or five days ago, maybe longer.
13:56.011 --> 13:57.112
And so you see what it was.
13:57.152 --> 14:06.555
It was to get me into a hamster wheel where I might argue about Pfizer and what Nick said or didn't say, or maybe even say the words that I said out loud, which is Nick Hudson might be a liar.
14:06.595 --> 14:07.175
He might be.
14:09.836 --> 14:12.277
Everybody, Jessica Hockett might be a liar.
14:12.317 --> 14:13.057
It's possible.
14:14.110 --> 14:15.151
I would be devastated.
14:15.991 --> 14:17.372
Mark Kulak could be a liar.
14:17.412 --> 14:19.133
It's possible I would be devastated.
14:20.334 --> 14:26.778
But the idea that some anonymous guy from France who is complaining and calling me a crackpot with a crack pipe.
14:28.880 --> 14:34.103
It's hard for me to believe that this is a genuine guy who basically agrees with me on most things.
14:34.203 --> 14:39.627
But, you know, we're... I have a hard time speaking in English.
14:41.368 --> 14:42.789
But he's pretty good at tweeting in English.
14:46.735 --> 14:47.716
Look at his followers.
14:49.978 --> 14:58.685
Let's see, seven of nine is also one of these anonymous accounts that has been supportive at times, but as soon as I started to say clones and everything, nothing.
14:59.366 --> 15:01.148
Liam Sturgis doesn't follow me anymore.
15:01.228 --> 15:05.031
Jessica Rose, David Cartland, he's, I don't know.
15:05.511 --> 15:07.373
Peter McCullough, I don't know.
15:07.553 --> 15:08.914
Clucky I think is pretty good.
15:08.954 --> 15:11.176
Soothspider we have in the chat all the time.
15:12.457 --> 15:18.142
This I Am Spartacus guy is one of the dudes that was on Kevin McCarren and Charles Rixey's stream for a long time.
15:18.162 --> 15:23.025
So maybe he's just fans of Charles Rixey and Kevin McCarren.
15:24.006 --> 15:26.348
Brooke Jackson, I've met her a few times in person.
15:26.408 --> 15:28.069
Finderella, I'm not convinced about.
15:29.750 --> 15:32.072
This Cat PhD seems to be pretty cool.
15:32.432 --> 15:34.514
I really like this account, Robert Corgan.
15:34.934 --> 15:36.875
That guy seems to nail it a lot.
15:37.756 --> 15:39.758
I used to really like Janey Yaya.
15:40.838 --> 15:53.506
But she's she's I I think she's kind of not happy with me anymore, but I could be wrong I used to really like I used to see a lot of tweets by Bullwinkle moose, but I don't see them anymore Dayu is one of the guys that's in drastic.
15:54.306 --> 15:58.569
That's that's probably a bad person or at least somebody who's playing for them.
15:58.589 --> 16:09.596
There's Martin Neal There's coronavirus movie that's now called fall of Rome There's Michael Sanger Walter chestnut there's gets go that's a
16:10.444 --> 16:12.004
That's a panda person.
16:13.605 --> 16:14.145
Michael Singer.
16:14.185 --> 16:15.005
Oh, sorry, that was there.
16:15.565 --> 16:16.326
Laura Baden.
16:16.726 --> 16:17.146
I know her.
16:17.226 --> 16:18.226
I think she's from Canada.
16:19.307 --> 16:21.087
Steve Kirsch, Toby Rogers.
16:21.767 --> 16:25.828
So you see, it's this kind of illusion, right?
16:25.888 --> 16:31.130
If all of these people are following all of these anonymous accounts, and then these anonymous accounts are amplifying the
16:31.690 --> 16:43.096
the statements and the ideas and the people they want to, or amplifying the replies of some of these people to the people that they're trying to amplify, they create this little illusion where they can get away with doing this.
16:46.677 --> 16:53.561
By not following me, and only when I decided to invite him to the stream did he say to come on the stream, right?
16:53.621 --> 16:55.702
So, think about that for a second.
16:55.722 --> 16:57.783
We can all go back to Twitter and read it, he said.
16:58.823 --> 17:18.881
go back to Twitter and read it if you can penetrate the app and see that I asked anyone wants to come on they could come on and then this guy said he would and it was in the context of this and in the context of what I said which is they murdered people and pretended it was a gain-of-function virus and none of these people will say it
17:20.512 --> 17:21.693
And then he tweeted this.
17:21.813 --> 17:34.021
And so when he came on my stream, it's really interesting that he started out with a completely different tweet about Nick Hudson and Pfizer and wanted to explain how Sasha Latipov was wrong because Pfizer is bad or she's wrong about Pfizer.
17:34.121 --> 17:38.304
And isn't it interesting that we have to talk about Sasha and Nick now?
17:39.144 --> 17:44.928
Sasha and Nick, the same two that had a Twitter space a couple of days ago and didn't invite me, right?
17:46.138 --> 17:47.879
And so it's fine, I don't care.
17:48.139 --> 17:56.003
If Nick Hudson wants to promote my Wolfgang Wodach interview as one of the best in the pandemic, I'm fine then.
17:56.524 --> 17:58.905
Whatever else he does, he doesn't always see my tweets.
17:59.005 --> 18:00.526
I don't think Nick Hudson's a bad guy.
18:00.546 --> 18:02.627
He might make mistakes, but I don't think he's a bad guy.
18:04.471 --> 18:10.016
And the mice are really desperate for me to think that Nick is a bad guy, but Nick has talked to me in private.
18:10.676 --> 18:13.279
Nick uses his name since the beginning of the pandemic.
18:13.799 --> 18:17.462
Nick has argued with me and then conceded many, many times.
18:17.942 --> 18:27.851
Nick has communicated with me about things that he's gonna do and people he's gonna meet and ask for questions and then reported back and has seemed to have his mind changed a few times about certain people.
18:28.551 --> 18:29.732
I have a long history
18:30.673 --> 18:57.714
Nick being a pretty trustworthy guy most importantly because he promotes my work Retweets my tweets, which is nothing like putting a crack pipe and saying he's running short and getting unhinged again from France But it's an important data point ladies and gentlemen, it's an important data point because that's how These people are misleading the young that's why they're doing it.
18:57.754 --> 18:59.115
That's how they're doing it.
18:59.135 --> 18:59.155
I
19:05.292 --> 19:05.752
The End
19:15.509 --> 19:18.170
And so if you want to go back and see it, you can go back and see it.
19:18.351 --> 19:21.152
It's definitely Cancelled Mouse and Anonymous Critic.
19:21.272 --> 19:24.894
It's not my best hour because I get really angry when people are lying to me.
19:25.915 --> 19:27.516
And I get a real sense for it too.
19:27.956 --> 19:34.099
And so there was a gut feeling that came over me at some point when I realized that
19:36.423 --> 19:39.086
That he was being disingenuous is the best way I can say it.
19:39.106 --> 19:40.688
So we're going to skip that one.
19:41.869 --> 19:46.674
I'm going to remind you of this one more time and then we're going to go on to this one, which I really like a lot.
19:46.794 --> 19:52.580
So let me just get that open on the Maccy Mac over here.
19:57.243 --> 20:00.025
And again, we're going to have the same problem as we always do.
20:00.085 --> 20:00.805
It is the Mac.
20:01.265 --> 20:07.769
It's a little slow and it will choke on YouTube ads, of course.
20:08.089 --> 20:08.930
Why is it not running?
20:12.152 --> 20:13.532
I hope everybody is okay.
20:13.652 --> 20:14.073
I hope you...
20:18.945 --> 20:23.129
Jicky told me that Jessica Hawk is a bad actor and that she's destroying Panda.
20:23.149 --> 20:23.869
That's hilarious.
20:23.910 --> 20:26.091
Because that guy was saying that Panda is a Psyop.
20:26.692 --> 20:28.173
A Psyop, which is pretty funny.
20:28.414 --> 20:29.074
Um, anyway.
20:29.655 --> 20:30.676
Let's go here.
20:31.056 --> 20:32.978
I don't know why this is not loading very quickly.
20:34.079 --> 20:34.799
But it will load.
20:36.321 --> 20:38.262
And then the history.
20:39.363 --> 20:40.324
Wow, that is slow.
20:40.885 --> 20:42.226
I might have to open the door or something.
20:44.965 --> 21:11.881
So we're gonna do this one first and then I'm gonna watch the next one after that I don't know we'll get through this one because it's only 15 minutes So the point I'm gonna try and make here is The thing I'm gonna try and point I'm gonna try and make with this video is that AI and its ability to solve the protein folding problem has been over exaggerated has been exaggerated to the nth degree very much like how
21:14.093 --> 21:15.222
Very much like how, uh...
21:20.290 --> 21:32.458
very much like how Peter Thiel described it on the Portal podcast in 2019, right before the pandemic, where he said that the hyper-specialization has allowed people to lie and exaggerate, and that's why he starts with skepticism all the time.
21:33.459 --> 21:45.387
And so, I think that AI is being exaggerated in this way, and being used as an excuse why we kill people in Israel, being used as an excuse why people get on lists, why people get censored,
21:46.527 --> 21:48.248
you know, being used as an excuse.
21:48.268 --> 21:48.408
Why?
21:48.448 --> 21:50.589
Well, we don't really know how the algorithm works.
21:51.529 --> 21:58.772
Being used as an excuse to select drugs like remdesivir and ivermectin and silicoxib and famotidine.
21:59.973 --> 22:00.913
These are all lies.
22:01.053 --> 22:12.418
And so if we go all the way from the domain server all the way up until this talk, where he's going to talk about protein structure predicting, you're going to see him hand waving because he wants to believe
22:12.938 --> 22:21.966
that these people aren't charlatans, but the next video that we'll watch should demonstrate pretty concretely that these people are indeed charlatans.
22:22.026 --> 22:23.628
So, let's do this!
22:26.571 --> 22:28.312
Oh, that's the wrong mouse there, big guy.
22:29.133 --> 22:32.396
Hello, everyone.
22:32.756 --> 22:33.797
My name is Michael Levitt.
22:34.638 --> 22:36.460
I'm a professor at Stanford Medical School.
22:37.316 --> 22:41.159
in structural biology with an association also in computer science.
22:41.900 --> 22:52.789
And today, I want to talk about protein folding, structure prediction, and biomedicine, three seemingly unrelated subjects that are actually very connected in this current world.
22:53.429 --> 22:56.572
But first, a slide on the secret of life.
22:56.752 --> 23:01.395
And basically, the secret of life is learning and self-assembly.
23:02.336 --> 23:05.597
For thousands of years, we've been looking for the secret of life.
23:06.118 --> 23:08.398
We now know it, these two things, and learning.
23:09.079 --> 23:11.200
So here is all of life in a single slide.
23:11.680 --> 23:13.961
The DNA structure contains the information.
23:14.561 --> 23:20.823
That information is coded from information into a physical object, the protein structure.
23:20.843 --> 23:22.984
The protein structure is precise.
23:23.064 --> 23:24.545
He's already revealed it, right?
23:24.745 --> 23:29.407
The main thing there is the ribosome.
23:30.629 --> 23:33.470
He's not even talking about the RNA intermediate here.
23:33.530 --> 23:51.239
He's telling you that in this very simple summary of the secret of life, it is DNA sequences turning to protein structures, and if those protein structures suck, then somehow or another, that error has to be back-propagated to the genome, or that genome has to be eliminated.
23:51.520 --> 23:52.540
Let's listen carefully.
23:53.140 --> 23:53.761
It's impressive.
23:54.710 --> 23:56.551
to a tenth of a nanometer scale.
23:56.631 --> 24:01.294
So it's more precise than any printed circuit, probably by a factor of a thousand.
24:01.895 --> 24:09.700
That structure then has a function which is based on physical interactions of, in this case, a red drug molecule and the structure.
24:10.220 --> 24:12.462
If this interaction is good for us,
24:12.782 --> 24:33.392
And so now one of the things that I want to bring in right here is that what's missing from this, of course, is that this interaction with the protein, the folding of the protein, and the interaction with the drug all occurs not between the protein and the drug, but between the protein, the drug, and the water that is surrounding these.
24:34.273 --> 24:36.894
And the water can be organized in very different ways.
24:37.014 --> 24:40.396
And the way that the water is organized is based on its extreme polarity.
24:41.553 --> 24:50.255
And so they come together a little bit predictably and a little bit crystalline, especially at the size scale of proteins, especially around a charged protein.
24:50.275 --> 24:54.056
They tend to organize the water around each other, around themselves.
24:54.556 --> 25:11.000
And the way they organize the water around themselves actually determines, I was about to reach over there and point, it determines exactly, sorry, I gotta get my mouse, exactly how this drug will interact with the protein because as they come together, the water will need to be extruded.
25:12.861 --> 25:13.902
He's got to get out of the way.
25:14.303 --> 25:19.109
Or whatever water is there has to be there because it's kind of supposed to be there.
25:20.120 --> 25:46.845
And so it's interesting that what is one thing that's very consistent about all these people that we've ever watched, whether it's David Baltimore or that lady from MIT that's now dead, whenever they talk about proteins and protein folding, they leave out the fact that it's inside of a polar solvent that organizes itself on a molecular level in a crystalline lattice and that these things are polarized and therefore they affect the way that crystalline lattice manifests around themselves
25:47.225 --> 25:51.850
And when they come closer to one another, the way that they affect that manifest then is how they interact.
25:53.292 --> 25:55.193
Have you ever heard a biologist say that to you?
25:56.175 --> 25:57.896
I've never heard a biologist say it to me.
25:57.936 --> 26:06.125
Not one teacher that I've ever had in my history has ever said it, yet I know it in my bones that that's what's missing from this picture.
26:11.578 --> 26:15.865
You learn by evolution, and evolution is a form of backpropagation.
26:16.265 --> 26:19.811
If there's errors in this, you die and you don't learn.
26:20.580 --> 26:33.893
And so that's kind of annoying because backpropagation is a very specific phenomenon in neuronal learning that is thought to be related to the membrane currents and how they communicate in time.
26:34.814 --> 26:42.462
And so calling this backpropagation, in my humble opinion, is an imprecision that's not helpful.
26:43.746 --> 26:45.447
So here's just showing you what the protein is.
26:45.527 --> 26:47.488
Proteins are really miraculous things.
26:47.828 --> 26:53.432
They're a material that make life simply possible.
26:53.772 --> 26:59.716
Proteins are a long chain of 20 different amino acids, like a necklace of 20 different colored stones.
27:00.376 --> 27:06.537
But that chain folds up, and you can see that the unfolded chain is much, much bigger than the folded chain.
27:06.977 --> 27:15.459
And the folded chain, in this case, a small protein with only 56 amino acids, the folded chain has a very precise three-dimensional structure.
27:15.979 --> 27:19.940
This is a protein that is an inhibitor of the enzyme trypsin.
27:20.500 --> 27:21.561
It exists in all of us.
27:22.001 --> 27:24.181
It also exists in almost all mammals.
27:24.541 --> 27:27.382
So basically, this shape has a very specific function.
27:29.571 --> 27:40.745
Here I'm just showing the same protein in two different views, showing you the colored beads, and then showing you that each amino acid, each vertex is an atom, also has a very precise shape.
27:41.246 --> 27:45.852
So protein folding is like making a three-dimensional jigsaw puzzle.
27:48.018 --> 27:55.707
OK, now one really important thing to remind you about proteins is that this process of folding up is completely spontaneous.
27:56.768 --> 28:02.014
It's interesting, but I guess kind of also obvious, that life has to be self-assembling.
28:02.815 --> 28:06.179
We're used to everything we know about being assembled from the outside.
28:07.000 --> 28:08.582
But in life, there was no aside.
28:08.642 --> 28:10.866
It had to be built up from first principles.
28:11.487 --> 28:15.833
And this is something which I think in the future, manufacturing will have to learn from.
28:15.873 --> 28:18.537
Biology is full of really, really good lessons.
28:20.282 --> 28:25.104
Of course, this is what people talk about when they talk about nanomachines and cellular biology.
28:25.124 --> 28:27.625
They're really interested in how these nanomachines work.
28:27.685 --> 28:33.588
Not enzymes, but things like a ribosome, which is a combination of ribonucleotides and proteins.
28:33.628 --> 28:36.950
It's a very, very complicated machine that we don't know jack about.
28:37.964 --> 28:47.487
Now, just moving forward to multiscale modeling of macromolecules, this is an area that I have been working in now for more than 50 years.
28:47.667 --> 28:55.989
I started doing independent research in 1967, and this research was actually recognized by the Nobel Foundation.
28:57.049 --> 29:01.690
The person who really got everything started was an Israeli scientist, Schneer Lifson.
29:04.031 --> 29:10.876
which now lives in students, and I came to Israel in 1967 to be a programmer for Lifson and Varshall.
29:11.416 --> 29:16.599
Martin Kaplis visited Israel afterwards, was very enamored by the idea.
29:16.619 --> 29:19.081
I re-edited a post-doc with Martin Kaplis.
29:20.162 --> 29:25.986
So I guess it's very sad that you know, Lifson passed away before the prize in 2013.
29:26.666 --> 29:28.267
But it would have been very difficult to decide.
29:28.687 --> 29:30.369
Nobel prizes are only for three people.
29:30.389 --> 29:32.830
It would have been very hard to decide who was left out.
29:34.271 --> 29:34.872
Everything is fine.
29:36.714 --> 29:46.244
And basically, one thing that we did very early on in 1975 was a paper on simulating protein folding.
29:46.864 --> 29:54.452
Essentially, we were able to take a protein chain and simplify it to be a chain of beads, like you would imagine it to be.
29:54.492 --> 29:56.914
But the important thing being the chain property.
29:57.635 --> 30:08.382
And then using energy minimization and something we call normal mode thermalization, we could take the long chain and get it to spontaneously fold up.
30:08.442 --> 30:18.469
Now, this whole thing took about 20 minutes of computer time on a computer circa 1975, quite quick.
30:19.129 --> 30:20.210
It wasn't very accurate.
30:20.370 --> 30:24.052
This structure is folded, but it's not exactly the right answer.
30:25.513 --> 30:29.517
Probably today you would say it has a GDT score maybe of 10 or 50.
30:29.997 --> 30:34.702
It's still similar, but not really that close, an RMS deviation of 5 angstroms.
30:35.623 --> 30:37.324
But the important thing was the concept.
30:37.985 --> 30:46.553
And it's also really important to realize that since then, computers have increased in speed by something like 10 to the 9, 1,000 times 1,000 times 1,000.
30:50.356 --> 30:52.399
This has made a lot of things possible.
30:52.519 --> 30:58.446
People often think that AI is possible because of new discoveries and algorithms.
30:58.887 --> 31:08.819
That's partially the case, but the main thing that has made computers so prevalent in our lives is the speed of computing, which has just increased in an incredible way.
31:09.713 --> 31:20.038
Okay, so now I want to tell you about three projects that relate more modernly, jumping forward in time from 1975 to the present.
31:20.698 --> 31:34.064
One is a project that has been done by colleagues of mine at Baylor called Opus X. And this is a project trying to predict accurate torsion angles using neural networks.
31:35.105 --> 31:36.806
Like neural network models,
31:37.782 --> 31:43.226
there are almost always many boxes that are connected together by networks.
31:43.906 --> 31:51.652
And this is able, again, using many different features taken together to get a structure which looks quite good.
31:51.992 --> 31:53.333
This is done by a small team.
31:54.614 --> 31:56.415
Here's a movie showing their result.
31:57.376 --> 32:03.442
The correct answer is the blue structure, and this structure starts out very disordered and then folds up.
32:03.943 --> 32:06.245
So that is an example of protein folding.
32:06.825 --> 32:13.131
It's a large protein, and my guess is this is a... So it's interesting to think about, just to let you know what I'm thinking about,
32:13.952 --> 32:16.274
I'm thinking about the process of making a protein.
32:16.314 --> 32:20.997
So does it come out of the ribosome and fold as it's being translated?
32:21.057 --> 32:28.063
Because then the way that they model it as a string that then folds to lower energy is fundamentally wrong.
32:28.103 --> 32:35.608
The model of how it happens would be wrong unless that comes out as a string and then once it's out of the ribosome, it starts to fold.
32:35.768 --> 32:36.309
Otherwise,
32:37.301 --> 32:42.766
they should be making the model, assuming that it folds as it's produced.
32:42.906 --> 32:49.113
And that might actually change the way that the prediction arrives at the endpoint, but they don't do that, right?
32:49.173 --> 32:57.280
So there's all kinds of aspects about this that I imagine if we were in the audience, we could ask questions and it would basically reveal
32:57.881 --> 33:08.943
that so many assumptions and so many corners are cut in order to get this, that applying it to a novel protein or a protein with a lot more subunits than this is just absurd.
33:10.344 --> 33:12.364
Again, it's brute force computing, by the way.
33:13.044 --> 33:26.007
And that's part of the argument that I make on this slide over here, is that they really believe... Oh, I gotta resume slideshow.
33:26.167 --> 33:26.787
Oh no, that's the...
33:28.126 --> 33:30.507
So I can escape and then I'm going to go.
33:30.928 --> 33:32.709
That's what I mean by this slide here.
33:33.950 --> 33:40.013
Um, they all believe that if we, what's going to happen, that's what Kurtzweiler is also arguing, right?
33:40.033 --> 33:40.894
Where's the slide?
33:40.954 --> 33:41.234
Dang it.
33:44.830 --> 33:51.875
They are arguing that once we get enough computing power, then our AIs will be able to do things they can't do right now.
33:52.575 --> 33:56.618
And actually, Leavitt is kind of making that argument that it's not algorithms.
33:56.698 --> 34:05.184
We're using the same simple neuronal network algorithms to, and we're just making more and more and more layers and putting more and more processors on it.
34:05.524 --> 34:11.668
We got a whole processor farm on the, on the, on the plains of the Netherlands with a windmill power filling it.
34:11.728 --> 34:14.690
So, you know, we can just, we'll just brute force it.
34:16.331 --> 34:20.933
And I would argue that brute-forcing the irreducible complexity of biology is a pipe dream.
34:21.093 --> 34:30.817
It's something that these people, like Raymond Kurzweiler, that says that within 10 years disease will be gone and you'll be able to basically be immortal, is this just nonsense.
34:31.397 --> 34:32.217
It's nonsense.
34:32.977 --> 34:37.859
And more importantly, it's an abomination to everything that our children represent as pure beings.
34:38.620 --> 34:39.180
You know, born
34:41.983 --> 34:42.664
Born perfect.
34:44.686 --> 34:46.828
So yeah, sorry, I didn't need to do that.
34:46.868 --> 34:47.889
I just needed to do this.
34:48.610 --> 34:50.051
And that should be this one and that one.
34:50.352 --> 34:50.992
And then we'll go.
34:52.274 --> 34:53.635
Selected good cases.
34:53.875 --> 34:56.298
Doesn't always work like this.
34:57.639 --> 35:00.462
But in some ways, this is a group that's been working.
35:01.183 --> 35:03.766
The chief person on this is
35:06.297 --> 35:09.479
Mao Zedong, he's been working in this field for 30 years.
35:09.519 --> 35:12.141
I've known him for a very, very long time.
35:13.722 --> 35:31.332
Another discovery, which is much more in the news, is that DeepMind, which is a wholly owned subsidiary, actually not of Google, but of Alphabet, the parent company that also owns Google, has been doing a lot of alpha things, alpha chess, alpha go.
35:31.372 --> 35:35.395
And then they decided to apply their machine learning experience
35:36.075 --> 35:37.061
to folding proteins.
35:38.223 --> 35:40.564
And I think for them, they had a very large team.
35:41.644 --> 35:44.925
They do have some people on the team who actually know about proteins.
35:45.625 --> 35:53.307
But generally, it's been an effort in getting into a new field without a lot of domain expertise.
35:54.107 --> 35:57.568
And they did really, really well.
35:58.848 --> 36:00.408
This is their score of 240.
36:00.508 --> 36:06.370
This is the second best score, which is almost exactly the same as the third best score.
36:07.230 --> 36:17.119
This is their team, this is a team led by David Baker, and this is a team led by three people, Jiang, et cetera, at the University of Michigan in Ann Arbor.
36:17.880 --> 36:19.782
And then the others pay it off.
36:20.242 --> 36:22.965
And these are, so they did very, very well.
36:24.126 --> 36:28.130
But again, it's important to realize that this is a blind competition.
36:29.151 --> 36:30.312
A sequence is released.
36:31.434 --> 36:35.095
And the group has three or four weeks to predict the structure.
36:35.575 --> 36:37.016
But not just one straight sequence.
36:37.376 --> 36:40.917
In this period, maybe 40 or 50 sequences are released.
36:42.077 --> 36:45.078
And this is a competition a little bit like the Olympic Games.
36:45.458 --> 36:46.238
Scale matters.
36:46.578 --> 36:47.859
If you have many athletes,
36:48.519 --> 36:51.760
you'll get more medals than if you have fewer athletes.
36:51.780 --> 36:54.781
So maybe the score needs to be somewhat normalized.
36:56.002 --> 37:00.884
This shows the black line, how well DeepMind did.
37:01.984 --> 37:04.725
And you can see here, these are targets that are easy.
37:04.765 --> 37:11.848
An easy target is a structure that is quite similar to a structure in sequence to a structure that's already known.
37:12.708 --> 37:20.055
So DeepMind does quite well, or maybe better than others, but not necessarily the very best on the easy targets.
37:20.696 --> 37:26.221
But when you come to the very difficult targets, where you don't have a lot of structure, they do very well.
37:26.261 --> 37:29.604
So for this, this was the thing that really surprised me.
37:29.624 --> 37:31.185
But I think there are two surprising things here.
37:31.305 --> 37:32.206
One is that it's a group.
37:33.027 --> 37:35.690
that doesn't have domain expertise, a lot of it.
37:36.311 --> 37:44.720
And it's important to realize that domain expertise is actually important, but their work had been preceded by 60 years of research in this field.
37:47.102 --> 37:48.444
Their models were interesting.
37:48.464 --> 37:59.616
They used convolution networks, recurrent networks, graph networks, where they actually connected together the learning units in the same way that atoms are close to each other in proteins.
38:00.237 --> 38:01.058
They were adaptive.
38:01.538 --> 38:03.520
And then sort of an attention model.
38:03.560 --> 38:06.804
Again, this group knows AI very, very well.
38:08.445 --> 38:10.686
They did something which I thought was interesting.
38:10.826 --> 38:18.950
And in some ways, it's exactly the opposite of what I did back in- Correct me if I'm wrong in the chat, but I would say that he's using AI and machine learning incorrectly.
38:19.010 --> 38:26.993
He's using them interchangeably when what he showed in the previous slide, I think would be considered, this is deep learning.
38:27.033 --> 38:28.134
This is machine learning.
38:28.414 --> 38:29.635
This is not AI.
38:32.876 --> 38:33.476
I don't know how
38:35.868 --> 38:39.349
how that really works for computer people.
38:39.369 --> 38:41.289
But I think he's using that incorrectly.
38:41.309 --> 38:42.329
This is just machine learning.
38:42.890 --> 38:46.430
Model again, this group knows AI very, very well.
38:48.051 --> 38:50.291
They did something which I thought was interesting.
38:50.451 --> 38:52.172
And in some ways, it's actually opposite of what I did.
38:52.312 --> 38:52.632
There you go.
38:52.712 --> 38:54.932
OK, so Freeman says that AI doesn't exist.
38:54.972 --> 38:55.833
It's all machine learning.
38:55.853 --> 38:56.793
Well, that makes sense to me.
38:56.833 --> 38:57.433
That rings well.
38:58.193 --> 39:01.234
In 1975, they basically said there is a chain.
39:02.537 --> 39:08.498
But let's break the chain up into rigid pieces and treat the whole protein like a three-dimensional jigsaw puzzle.
39:09.259 --> 39:14.960
And the good thing about doing that is that you could treat this whole big piece as a single rigid unit.
39:15.380 --> 39:20.941
So you can break anything up into rigid units and then have something favoring connecting them.
39:21.321 --> 39:23.642
So basically, it did very, very well.
39:24.916 --> 39:38.046
But it's also very important to realize that, like all science, it's very well based on previous work, but even more so, because machine learning needs examples, and the more examples, the better.
39:38.566 --> 39:42.809
So they had hundreds of thousands of structures from crystallographers,
39:44.191 --> 39:47.133
NMR specialists, cryo-EM people, to solve the structures.
39:47.753 --> 39:53.958
Then they needed, for each of their structures, they often had 500 or 600 different sequences from different species.
39:54.458 --> 40:01.344
The idea being that these variations should tell you what are the positions that matter and what are the positions that don't matter.
40:02.104 --> 40:09.310
And a lot of previous work in terms of representations, but it's still a very important finding.
40:10.431 --> 40:14.476
So determining structures is very important, but in fact, it's almost like a game.
40:14.496 --> 40:21.205
I mean, playing Go is very important, but it's not really important for well-being.
40:21.626 --> 40:25.050
What is really important is what you do with the protein structures.
40:25.959 --> 40:31.365
And one thing that we are all very concerned about is getting better pharmaceutical agents.
40:32.306 --> 40:37.732
These are small molecules which, given to us in a pill, can actually make us better.
40:37.912 --> 40:41.636
And human health is obviously a problem of central concern.
40:42.117 --> 40:43.719
Structural prediction, maybe not.
40:43.759 --> 40:45.120
So biomedicine is critical.
40:46.161 --> 40:55.094
So basically, I want to just mention a company run by Alex Zhaborodtsov, and it's a startup company.
40:55.975 --> 41:00.602
And basically, like the work on Opus X, they have many boxes that they connect.
41:02.113 --> 41:06.054
And basically, these are connected together using AI.
41:06.694 --> 41:09.534
I like this picture because it shows these as if they're printed circuits.
41:10.155 --> 41:23.037
But basically, one thing I would say about in silico medicine is they use AI for target discovery, for chemistry, for virtual screening, for clinical trial outcome prediction, and my guess is for anything else.
41:23.317 --> 41:28.638
They are literally AI from A to Z. AI is very much the core of the company.
41:29.658 --> 41:31.721
So this is a very important thing to realize.
41:32.101 --> 41:37.888
And the idea is to use AI not for one component, but for the entire pipeline.
41:38.148 --> 41:42.494
Pharmaceutical chemistry, drug discovery is a very wide project.
41:42.574 --> 41:44.296
It goes from protein structure
41:44.916 --> 41:52.098
through chemistry, through screening, with toxicity, with clinical trials, many, many components.
41:52.538 --> 41:58.039
And AI is actually particularly good at integrating input from many components.
41:58.079 --> 42:08.522
Now, what I find most interesting about this is that I think that, if we go back just a few, I think that isn't this what the domain server already does?
42:11.465 --> 42:22.790
Robert Malone said the domain program with DITRA was actually an AI computer simulation of X-ray crystallography of the three CL protease.
42:23.530 --> 42:36.776
And then that AI took and fit all kinds of the entire catalog of the FDA pharmaceuticals and nutraceuticals and identified four drugs that were good candidates for antivirals.
42:37.536 --> 42:40.918
And then he proceeded to start testing them in combination.
42:43.806 --> 42:51.252
So I'm confused, is Robert Malone very, very much ahead of where this, is he ahead of this?
42:52.413 --> 42:53.634
Or what's happening here?
42:54.415 --> 43:08.186
Because it seems like he's not aware that DITRA already has a program that can do this, has already successfully scanned the entire FDA catalog and identified four things in a matter of weeks with a volunteer team.
43:09.844 --> 43:14.426
So it's a little weird that he's like plugging Google if it's a three-man team and they did really well.
43:14.846 --> 43:15.587
Holy shit!
43:16.147 --> 43:21.390
Robert Malone and David Hohn kicked some ass in February of 2020 then.
43:21.930 --> 43:25.812
And Michael Levitt doesn't seem to be aware of how much of a hero they actually were.
43:28.029 --> 43:28.769
And I'm impressed.
43:46.676 --> 43:48.518
And they're basically using this technique.
43:48.538 --> 43:49.939
They're trying very, very hard.
43:50.020 --> 43:50.880
It's a philosophy.
43:50.920 --> 43:52.942
It doesn't need to be the correct philosophy.
43:52.983 --> 43:55.345
It doesn't need to be a philosophy that's going to succeed.
43:55.405 --> 43:57.567
It's a mindset.
43:58.068 --> 44:00.210
Exactly how I would say it.
44:00.350 --> 44:01.891
It is a mindset.
44:02.032 --> 44:03.353
They have a mindset.
44:03.613 --> 44:04.834
They have a mindset.
44:05.035 --> 44:05.955
What is the mindset?
44:06.556 --> 44:10.101
They think that the AI is going to come to their rescue.
44:10.141 --> 44:12.084
It's going to solve all of their problems.
44:12.665 --> 44:18.133
It's going to, with enough brute force, it's going to solve the irreducible complexity of humanity.
44:19.517 --> 44:29.721
to speed up the traditional approaches where it takes hundreds of millions of dollars and many years to even get to the stage of elite optimization.
44:29.761 --> 44:32.823
They're trying to reduce their time dramatically.
44:33.523 --> 44:34.863
The millions become thousands.
44:35.224 --> 44:36.864
So it's a very dramatic change.
44:37.224 --> 44:43.607
My prediction is that if Michael Levitt did anything during the pandemic, he made sure to support the faith.
44:44.888 --> 44:47.449
Faith in a novel virus, the possibility of a lab leak,
44:49.366 --> 44:49.946
I'm sure of it.
44:50.346 --> 45:00.670
Maybe he didn't, maybe he argued for natural virus, but I'm sure that he did not question what happened and did not question something strange was going on.
45:01.350 --> 45:04.991
And with this, they've actually been really quite successful.
45:05.472 --> 45:06.492
This is the hard step.
45:07.659 --> 45:08.600
How do you find a target?
45:08.840 --> 45:09.861
Pandaomics.
45:09.981 --> 45:15.747
And what they did, and I think it's really interesting, is look at profiles of many people.
45:16.367 --> 45:18.229
But how do you say... Pandaomics, oh my goodness.
45:18.269 --> 45:21.972
If you know what the disease is, you can say sick people or healthy people.
45:22.453 --> 45:24.355
They did something which I think is very clever.
45:25.135 --> 45:28.238
Old people generally have more wrong with them
45:28.979 --> 45:34.043
than young people, in the same way that old motor cars have more wrong with them than new motor cars.
45:34.323 --> 45:48.714
But the things that are wrong with an old person will be reflected in their profile, in their biome, in their epigenetics, maybe not in their DNA, but in which proteins are expressed.
45:49.214 --> 45:55.579
So by looking comparatively at things that are different between young and old, you start to be able to identify diseases
45:55.839 --> 46:00.561
Is he asking for a proteome screen of everybody as they age?
46:01.862 --> 46:02.742
Sounds like it to me.
46:05.523 --> 46:08.665
Sounds like he needs a lot more data than he can get from a database.
46:10.986 --> 46:22.251
Sounds like he's advocating for everything that we have been saying they are trying to do, which is to start to use us like experimental animals and extract the data that they need to feed their AI.
46:23.257 --> 46:27.840
So in fact, in silico medicine, it's taken the pipeline even further.
46:27.880 --> 46:30.401
We now have end enabling as part of the thing.
46:30.921 --> 46:39.046
And very importantly, in idiopathic fibrosis and kidney fibrosis, we're actually at a stage now of starting microdosing in humans.
46:39.346 --> 46:40.967
So I think this is the way to go.
46:41.367 --> 46:46.871
The idea being that by going from A to Z with AI, the whole process is speeded up.
46:48.151 --> 46:50.373
A lot of companies are working on lead optimization.
46:50.393 --> 46:51.794
What a thing to advocate for.
46:51.814 --> 46:53.195
What a thing to advocate for.
46:53.235 --> 46:54.576
Lead optimization, super good.
46:54.636 --> 46:56.097
And then you get stuck further down.
46:56.437 --> 47:04.063
And my guess is that they're going to be able to apply their methods further down the stream as they can to really optimize the whole process.
47:04.083 --> 47:09.627
So I think it's an example of using computer, using the ability to combine data.
47:10.168 --> 47:13.750
Also, in some cases, uncertainty is a good thing.
47:13.810 --> 47:15.512
By combining a lot of data,
47:16.072 --> 47:25.815
that's uncertain, AI is very good at coping with vast amount of data and certain techniques by combining them, coping with vast amount of data.
47:25.915 --> 47:26.616
Oh, nice edit.
47:26.656 --> 47:34.258
And certain techniques, by combining them with clever filtering, we can get certainty and options from uncertainty.
47:34.598 --> 47:38.120
The protein filtering problem, that's a very difficult problem for 50 years.
47:39.000 --> 47:41.581
Drug design, all being dealt with in this,
47:42.341 --> 47:44.184
global, all-encompassing way.
47:44.805 --> 47:52.418
One of the things that I think is a red thread running through this is that everybody is sure that we're confused, and they're also really sure about why.
47:55.677 --> 48:01.580
This guy's really sure that eventually there's gonna be enough computing power so that all of these assumptions will be met.
48:02.161 --> 48:03.641
All of these goals will happen.
48:04.162 --> 48:06.503
They will be able to use AI for everything.
48:06.523 --> 48:08.144
They won't have to do testing anymore.
48:08.624 --> 48:22.712
Don't you see, that's where, that's exactly where personalized medicine is, where they don't do testing anymore because it's everything designed for you based on your biological signatures, your protein expression, exactly what he argued for right here.
48:24.724 --> 48:26.525
And this is during the pandemic.
48:27.465 --> 48:29.206
This is like in 2021 or 2022.
48:29.546 --> 48:33.128
This is extraordinary.
48:34.708 --> 48:35.709
Very, very optimistic.
48:35.769 --> 48:37.550
So I just wanted to thank you all for attention.
48:38.050 --> 48:39.771
And I think we have a great future to look forward to.
48:39.811 --> 48:40.151
Thank you.
48:40.351 --> 48:44.513
A great future to look forward to.
48:45.193 --> 48:49.515
Let's pause that and then let's go to the history.
48:51.343 --> 48:54.084
and let's get this other one up, and then I'm gonna go to basketball.
48:54.104 --> 48:57.325
So the boys are gonna go to basketball now, but I'm gonna meet them there so I can play later.
48:58.525 --> 49:03.327
This is really great, because I don't know how long we have to go, but this one is a really good one.
49:05.428 --> 49:08.449
This is how AI optimize.
49:08.469 --> 49:12.510
AI, AI, AI, they want everybody to believe that AI is something magic.
49:26.710 --> 49:29.592
discovery, AI and drug discovery.
49:30.413 --> 49:33.015
And it will be a very controversial presentation.
49:33.095 --> 49:37.998
So be nice to me and I will try to be as nice as possible to you as well.
49:40.080 --> 49:48.526
Before I introduce you to the workflow of the whole presentation of the points, I would like to set the framework.
49:49.406 --> 49:54.610
I would like to actually communicate the basic idea about AI and drug discovery to you.
49:55.496 --> 49:58.597
which is also kind of like a reflection of the status quo.
49:58.677 --> 50:11.139
So I've collected a lot of examples that I would go through to give you an idea of how impactful it can be, where the advantages are, and where also maybe pitfalls may appear here and then.
50:11.159 --> 50:19.801
Everything I present to you is a very subjective perspective from my side, because I am a medicinal chemist, and the judging is completely up to you.
50:20.001 --> 50:24.322
So sit down, lay back, and enjoy the whole session.
50:25.851 --> 50:30.133
The first topic we will tackle is actually what is machine learning and what is AI.
50:31.053 --> 50:39.618
Next, we will go through how to collect data for your approaches and also different methods in machine learning and artificial intelligence.
50:41.138 --> 50:45.060
The next thing is actually the drug discovery done by machines.
50:45.100 --> 50:49.502
We will talk a little bit about the hype around it and the cool reality.
50:50.743 --> 50:54.405
And we will finish with drug discovery stories I have collected.
50:56.164 --> 51:03.729
also address the, I would say, problem of novelty in machine learning for drug discovery, and also how the industry adapts.
51:04.389 --> 51:07.931
Again, this is only a snapshot, also considering the publications.
51:08.491 --> 51:18.257
It's close to impossible to cover this whole area and the whole field with a few slides and one single presentation, given the fact that it's only supposed to last for maximum one hour.
51:19.598 --> 51:24.121
So take it with a grain of salt, as I would say, and yeah,
51:25.933 --> 51:32.995
The first thing I would like to address that AI and machining is without a doubt still a part of the computer-aided drug discovery field.
51:33.495 --> 51:44.359
So anything where you use machines, models, calculations to predict if a compound is of interest to you is a computer-aided drug discovery project.
51:45.639 --> 51:47.280
You can do a lot of things with that.
51:47.500 --> 51:54.802
I don't need to address every single point on this slide, but you have to keep in mind that machine learning is only a prediction and
51:55.583 --> 51:58.966
It can be good, it can be bad, but it's only a sole prediction.
52:01.287 --> 52:17.960
So since humanity was punished a few hours ago with the collapse of the Tower of Babel, we are now in an unfortunate situation where everybody speaks a different language, right?
52:18.020 --> 52:21.723
So there's English, there's German, there's French,
52:22.697 --> 52:30.060
And given the fact that even if we speak the same language, we sometimes mean different things while using the same or different terms.
52:30.801 --> 52:36.443
So AI and machine learning are usually used in the same manner.
52:36.523 --> 52:40.405
So if you just talk about, yeah, AI will revolutionize drug discovery.
52:41.425 --> 52:45.967
In ultimate theory, they also always or sometimes also referring to machine learning.
52:46.522 --> 52:59.072
So artificial intelligence is actually a broader concept where by the end of the day, the whole apparatus, so the whole engine, can come up with solutions on its own.
52:59.772 --> 53:06.357
Whereas machine learning is a subset of AI, so it can be kind of like an entry to the artificial intelligence.
53:07.178 --> 53:12.062
And machine learning applies different algorithms and statistical models to predict something, right?
53:12.523 --> 53:18.385
So we take data, we feed data to algorithms, and then we generate a model that is able to predict something.
53:18.925 --> 53:28.588
So again, even during this whole presentation, I will be showing to you now, if I use machine learning, I'm kind of like also referring to artificial intelligence and vice versa.
53:30.809 --> 53:40.452
So I think we all are now in a very exciting time zone, time age, where we are
53:40.885 --> 53:45.753
Getting in contact with AI can be applied on different areas.
53:46.174 --> 53:50.602
So I mean, who hasn't used chatty PT to rephrase a sentence for his thesis?
53:51.164 --> 53:52.664
or come up with synonyms.
53:53.024 --> 53:59.126
I have never, ever, ever used chatGTP or any AI chatbot.
53:59.206 --> 54:05.207
I have never used AI artwork generation, who also some people tried to get me into that a couple years ago.
54:05.748 --> 54:08.668
I've never bothered with any of it because I know it's bullshit.
54:09.688 --> 54:19.711
It's got to be just an illusion that they want you to believe if they get enough computing power, they're going to be able to sneak it in in the back and install it and say, see, there was always AI there.
54:20.512 --> 54:21.512
It's just ridiculous.
54:21.572 --> 54:24.173
It's machine learning and it's programmed to do stuff.
54:24.253 --> 54:24.653
That's it.
54:26.233 --> 54:30.875
Um, but it's also penetrating other fields like navigation with Google maps.
54:31.615 --> 54:32.995
Also your social media.
54:33.455 --> 54:37.937
Um, who would be calling me from Seattle, Washington?
54:38.477 --> 54:38.697
Thanks.
54:38.757 --> 54:39.417
Try to learn.
54:39.437 --> 54:41.037
Maybe it's Brett Weinstein.
54:42.038 --> 54:42.878
He's in Oregon, right?
54:42.898 --> 54:43.898
I don't know where he lives.
54:44.358 --> 54:46.579
Advertisements are getting more personalized.
54:46.639 --> 54:47.479
They can even predict.
54:47.998 --> 54:55.200
what is your favorite color and show the exact dress, and also even Netflix recommendations based on what you've been watching yesterday in the evening.
54:55.900 --> 55:04.903
So AI is now very prominent and it's maybe also a leading curse why it's also so prominent or becomes even more prominent in drug discovery.
55:05.684 --> 55:10.145
But the whole concept about AI is not so novel, at least for drug discovery.
55:10.814 --> 55:22.582
The first reports were in the 1950s, where basically the flavor of AI, so kind of like easing up the life, convenience increase in drug discovery has been pursued.
55:22.942 --> 55:38.913
And in 1965, the Stanford, University of Stanford applied the Dental Project, where they tried to construct an algorithm to read organic spectra, the mass spectra of organic molecules to predict their structure.
55:39.417 --> 55:41.138
What do you think he's talking about, Michael, right there?
55:41.698 --> 55:56.724
With the modern advances in machine learning and neural networks and computing, we can now see that this whole topic, researchers in the 21st century and the past, I would say, two decades, it becomes even more prominent.
55:56.904 --> 55:58.104
And it's becoming prominent.
55:58.204 --> 55:58.664
Why?
55:58.805 --> 56:07.388
Because of the same reason that Michael Levitt said, because computational power is starting to catch up in their imagination to what's required in order to crack the problems.
56:08.294 --> 56:18.820
They are only, they are not ascribing a progression in understanding that then can be used to leverage machine learning to greater understanding.
56:19.200 --> 56:29.505
They're talking about, we don't understand any more than we did 20 years ago, and we're hoping that if we put enough computers in a chain, that we'll be able to learn something, or the computers will show us what we missed.
56:30.366 --> 56:36.509
It's a very, he's doing very well, but you'll see, he's gonna be critical pretty quick.
56:38.134 --> 56:51.801
I think the computing, the increase in computing power is a major role because now every small group even has a small server farm standing somewhere in the corner that can process a lot of data and can be used to also store a lot of data.
56:52.561 --> 57:01.626
And also, yeah, the convenience of having access to a high-performance computing via Amazon Cloud and so on is also something that kind of like, yeah,
57:02.228 --> 57:15.622
nourished the whole idea of implementing AI and drug discovery because... One of the things that I have a problem with too is this idea of cloud computing and you're kind of just trusting that you're cloud computing, but what are you really doing?
57:15.662 --> 57:19.086
Is it just going to a room full of people that's analyzing whatever they want to look at?
57:19.126 --> 57:22.810
Is it just going to a big... So it's weird because again, you know,
57:24.286 --> 57:33.892
people are taking for granted that what they see on their screen is actually represented in the real world somewhere in a whizzing, spinning processing unit.
57:34.032 --> 57:44.558
When in reality, just like with Twitter, when you look on Twitter, you're not seeing an algorithmically generated random assortment of tweets based on your following and your previous behavior.
57:44.578 --> 57:49.581
You're seeing exactly what that algorithm wants to show you, has been programmed to show you.
57:51.564 --> 57:58.067
And that's why that guy canceled mouse can go around and try to do all the things that they do anonymously.
57:58.107 --> 58:04.991
Because again, they're programmed to rise, they're gaming the system because they probably have figured out or know how it works.
58:05.872 --> 58:07.592
And it doesn't have to be that guy in France.
58:07.633 --> 58:09.974
That guy in France could be a good guy, I don't know.
58:11.134 --> 58:14.596
But the vast majority of the mice that are out there are not good guys.
58:14.636 --> 58:16.117
And he admitted that several times.
58:16.517 --> 58:17.578
And so here we are again.
58:18.749 --> 58:20.971
We're being made to ask the wrong questions.
58:20.991 --> 58:29.638
We're being made to think that something is possible when it's not even assumed that it's possible because eventually there will be a lot more computers than there are now.
58:30.079 --> 58:34.463
And these guys are making the argument that if we had more computers that this would be possible.
58:34.983 --> 58:41.788
And I think that fundamental argument is very similar to if these people tweak viruses enough, then they're going to cause a pandemic.
58:41.848 --> 58:49.494
Even if we didn't have one this time from a lab leak, the next one could be, it's the same kind of, I don't really like it.
58:49.514 --> 58:54.117
A lot of data and requires a lot of computational resources.
58:55.438 --> 59:00.242
So, um, how does AI work?
59:01.460 --> 59:07.581
If you would completely boil it down to a very, very, very simple idea, it would look like this.
59:08.222 --> 59:10.362
The first thing you have is data.
59:11.242 --> 59:23.805
Then you transform this data, you use your transformed data to train a model, you check your model if it does what it's supposed to do, and then you make decisions based on the outcomes of the model.
59:25.365 --> 59:27.866
If we boil it down a little bit, it becomes a little bit more complicated.
59:29.021 --> 59:46.990
Now what we're supposed to be doing in science is we're supposed to be taking the data, we're supposed to be transforming it into figures so that people can understand what our working model of that little system is, and then we're supposed to predict the outcome of a new experiment based on our understanding of that model.
59:47.730 --> 59:53.133
And then that experiment will be done and we will either validate the model or we will discard the model.
59:54.674 --> 01:00:19.854
Here what they're doing is they're using that same logical process for starting to build understanding of the world and turning it over into three black boxes that the computer will do and so essentially what you're looking at here is a AI version of the scientific method and We're supposed to just surrender to it
01:00:21.118 --> 01:00:29.723
that it even works, that the programming of this, the transforming of this, the training of this is being done correctly, is all taken for granted here.
01:00:30.863 --> 01:00:34.185
Because if it's not done correctly, the first few iterations will do it correctly.
01:00:34.245 --> 01:00:35.046
Eventually...
01:00:36.380 --> 01:00:52.664
The AI and the computer power and all of the things that we need will come together in a singularity, like Ray Kurzweiler says, and immediately, you know, the AI will be visible and sensible and will all be uploadable and disease will disappear.
01:00:54.544 --> 01:00:56.325
I don't think this guy believes that's the case.
01:00:57.025 --> 01:00:59.525
I really like this term, data is the new oil.
01:00:59.645 --> 01:01:02.126
And I think it's kind of like really reflecting how
01:01:03.174 --> 01:01:06.695
Things have been data is the new oil.
01:01:06.955 --> 01:01:08.775
Oh, wow.
01:01:08.835 --> 01:01:09.695
That's a good one.
01:01:10.515 --> 01:01:23.318
If we're all doing in the past decade based on social media, where every single, uh, uh, word you use on social media can be used to predict your behavior, your consumer consume, uh, behavior.
01:01:23.338 --> 01:01:29.499
Um, so this is also where kind of like the whole things come thing comes together.
01:01:29.991 --> 01:01:33.053
Because the more data you have, the more you could actually extrapolate.
01:01:33.554 --> 01:01:35.996
But you don't need to process everything.
01:01:36.176 --> 01:01:40.099
So when you have the data, you need to collect this data, then you need to prepare the data.
01:01:40.119 --> 01:01:45.843
You need to select what do I want to focus on, which parts of the data are of interest to me.
01:01:45.883 --> 01:01:48.385
Do I would like to work only with active molecules?
01:01:48.505 --> 01:01:50.127
Then you need to do some sort of filtering.
01:01:51.147 --> 01:01:53.869
Once you have your data in place, you need to transform your data.
01:01:54.350 --> 01:01:54.950
What does it mean?
01:01:55.331 --> 01:01:58.333
Not everything that is data can be read.
01:01:58.977 --> 01:02:02.398
So let's assume you would have a picture of, I don't know, something huge, like a dog.
01:02:02.958 --> 01:02:06.560
So you have a dog picture and you would like to use the picture for machine learning.
01:02:07.120 --> 01:02:09.701
The picture is something that cannot be easily processed, right?
01:02:10.041 --> 01:02:16.223
So you need to transform it into something that can be read by programs algorithms.
01:02:16.903 --> 01:02:22.565
So given now the structure of SiliconShip at the bottom, we have different possibilities to represent the structure.
01:02:22.745 --> 01:02:26.786
So you could provide 2D data of that, but then you would actually do the same thing
01:02:27.253 --> 01:02:34.776
you would try to check for the 2D and process this image into something readable, like smiles or fingerprints.
01:02:35.536 --> 01:02:38.877
So fingerprints can be read because it's just a sequence of zeros and ones.
01:02:39.317 --> 01:02:49.461
Smiles, on the other hand, is also a readable format because it's just a sequence of numbers, of symbols, and also of letters.
01:02:50.672 --> 01:02:53.795
Once you have your data transformed, you use the data to train your model.
01:02:53.935 --> 01:02:57.298
So you choose a model or an algorithm you would like to apply on your data set.
01:02:58.179 --> 01:03:00.921
An important part is here to also split your data.
01:03:01.302 --> 01:03:13.353
So basically, data you would like to use to train your model, and then data that you are aware of the properties to use to check for the quality of the model subsequently once your model is finished.
01:03:14.485 --> 01:03:16.346
Then you model it.
01:03:16.386 --> 01:03:28.632
And so what's interesting about this is, as he's explaining how an AI would be used to do drug discovery, what did they split in the domain server?
01:03:28.712 --> 01:03:40.838
So remember, the domain server started with, according to Robert Malone, a computer-generated model of X-ray crystallography of the 3CL protease of the coronavirus.
01:03:41.929 --> 01:03:52.713
And this enzyme was the enzyme that he wanted the AI to find possible interacting targets with from the entire FDA database of pharmaceuticals and nutraceuticals that have been approved.
01:03:54.793 --> 01:04:04.777
And so he had the data from the FDA, because I guess the FDA had the chemical structures in three dimensions in water for all of these drugs.
01:04:05.876 --> 01:04:19.482
And then all of those models were interfaced with the model of the x-ray crystallography structure of the enzyme they were interested in blocking or interfering with antivirally.
01:04:20.743 --> 01:04:23.984
And so I'm wondering, for example, what algorithm they chose.
01:04:24.884 --> 01:04:33.248
Were they looking at a particular part of the enzyme, or where it would dock, or whether it would interfere with the way that the enzyme changes conformation during
01:04:36.280 --> 01:04:39.161
during RNA translation, it wasn't really clear to me.
01:04:40.161 --> 01:04:51.365
And it's certainly not clear to me now that I understand there are more details to this subtle process of designing an AI to screen a certain data set.
01:04:52.885 --> 01:04:57.386
And remember, Robert Malone was told by Michael Callahan to spin his team up, and then he did it.
01:04:57.686 --> 01:05:06.009
And in three weeks, they had screened the entire catalog of FDA drugs and pharmaceuticals with a novel AI program.
01:05:08.851 --> 01:05:09.931
Like that's pretty amazing.
01:05:09.971 --> 01:05:14.572
Cause he wasn't planning on doing it until Michael Callahan called him and said, you better spin his team up.
01:05:14.972 --> 01:05:17.153
And then he got volunteers to help him do it.
01:05:21.033 --> 01:05:21.173
Hmm.
01:05:21.654 --> 01:05:22.034
Let's see.
01:05:22.054 --> 01:05:27.455
I mean, I don't, I don't know what all these maybe domain did a lot of this stuff before the pandemic.
01:05:27.495 --> 01:05:28.535
Maybe they were already ready.
01:05:29.475 --> 01:05:30.775
That's whatever it's supposed to do.
01:05:30.975 --> 01:05:31.856
It learns patterns.
01:05:31.976 --> 01:05:33.796
And then finally a model is created.
01:05:35.942 --> 01:05:42.190
Once the model is created, the operator needs to assess and evaluate the prediction quality of the model.
01:05:42.651 --> 01:05:49.580
So you conform the model with unseen data, which can be the testing set, and also novel data.
01:05:50.100 --> 01:05:51.262
And then you assess the performance.
01:05:51.704 --> 01:05:58.167
Can my model predict actives from... I mean, did he not identify previously known antivirals?
01:05:58.447 --> 01:05:59.668
He didn't, apparently.
01:05:59.748 --> 01:06:15.036
I mean, it's very strange if you listen to how it's done and how it should happen, that Robert Malone's entry pass into this COVID narrative was that I work for DITRA and DoD, and we did the domain program.
01:06:18.325 --> 01:06:23.028
That's Robert Malone 2020 that nobody wants to talk about because the shots are bad.
01:06:25.290 --> 01:06:25.550
Or not.
01:06:26.911 --> 01:06:29.413
Finally, you need to make decisions.
01:06:29.533 --> 01:06:33.156
So you deploy the model for usage, you feed it with new data, and then
01:06:33.726 --> 01:06:34.587
actually use it.
01:06:34.947 --> 01:06:42.333
And this is where some companies struggle because, yeah, sure, you can construct a model, but you also need to get access to this model.
01:06:42.414 --> 01:06:43.835
So you actually... So it's interesting, right?
01:06:43.875 --> 01:06:47.998
Because all of these parts of this process should have been done in the domain server.
01:06:48.038 --> 01:07:01.150
And I think that if you wanted to really justify the selection of Silicoxib or Formotidine or Ivermectin or Remdesivir based on the results from the domain program, then some of these steps should be
01:07:01.730 --> 01:07:09.599
available to us and revealed in terms of methodological clarity so that we can see whether or not these recommendations were really useful.
01:07:09.639 --> 01:07:17.507
Maybe, you know, use the actual monitor or use the actual the actual model and have it identify previously identified
01:07:18.448 --> 01:07:19.009
substances.
01:07:19.049 --> 01:07:19.990
It's very strange.
01:07:20.430 --> 01:07:27.618
I think the domain server is probably just a big story that Robert Malone told because, again, that was kind of scripty.
01:07:28.418 --> 01:07:34.164
And now they're so off script that we can't go back to that part of the discussion anymore.
01:07:34.224 --> 01:07:37.688
And none of the anonymous accounts that are on the Twitter have
01:07:38.229 --> 01:07:50.261
including Jickie Leakes himself or herself, doesn't seem to be able to muster the courage to say that there are some American traders that are responsible for creating the illusion of a pandemic early on in 2020.
01:07:50.982 --> 01:08:00.832
And their names are Robert Malone, Meryl Nass, Jessica Rose, Harvey Reich, Corey, Pierre Corey, and many others.
01:08:04.588 --> 01:08:29.789
They may not be happy about where we are now, they may not be happy about what they've done, but they were co-opted, they participated, they eventually became witting and knowing participants in this lie, which actually started, according to Robert Malone, with an effort by DITRA and their domain program to identify repurposed drugs because Michael Callahan said that it was necessary.
01:08:30.922 --> 01:08:34.203
to use this model in an environment where it can't perform.
01:08:34.863 --> 01:08:43.966
It doesn't make sense to just construct a model and only you can use it, but if you can also include this model into an actual workflow.
01:08:44.126 --> 01:08:45.426
I mean, it would be great, right?
01:08:45.466 --> 01:08:56.549
If the domain server was in three weeks able to identify something as effective as ivermectin, which is now curing cancer, then shouldn't we put the domain server to work and find out some more?
01:08:59.414 --> 01:09:06.918
Maybe it can cure RSV, and measles, and polio, and all these other things that we didn't know we had medicine for because we didn't have domain.
01:09:09.140 --> 01:09:10.541
Stop lying!
01:09:12.182 --> 01:09:16.344
Access to this model, it greatly improves the performance of the model and also the outcomes.
01:09:16.364 --> 01:09:18.425
And finally, it makes your life better.
01:09:19.226 --> 01:09:22.047
Finally, you should also want to update your model.
01:09:22.608 --> 01:09:26.390
Every time you get the chance to actually feed it with new and improved data,
01:09:27.461 --> 01:09:30.803
use it because it can and will improve the quality of the model.
01:09:32.504 --> 01:09:38.788
The last three, sorry, the last four entries here are actually part of an iteration cycle.
01:09:39.168 --> 01:09:44.071
So if you are not satisfied with the performance of your model, it can always make sense to adapt your model.
01:09:44.431 --> 01:09:52.676
So you do some hyperparameter tuning, which means you adjust the ratings of connecting points, you adjust the data here and there.
01:09:53.257 --> 01:09:55.338
You can also feed new data if you say, well,
01:09:55.717 --> 01:10:01.395
I fed only 100 actives, maybe I should use more to also improve the
01:10:02.513 --> 01:10:28.132
So it would be really interesting to have Robert Malone explain the optimization steps that they took in order to tweak the domain server in order to get the results that they got because presumably there was a longer list and they just kind of cut it off at a certain relevance point or use their previous experience with the outbreaks of Zika and Ebola and all the other stuff that Robert Malone and his wife have been responding to for 20 years to have insight into how to tweak the AI
01:10:29.093 --> 01:10:34.517
processing of domain, but we don't have any details about that because we're not talking about 2020 anymore.
01:10:37.238 --> 01:10:41.681
Sometimes it can also make sense to see, well, my model is completely underperforming.
01:10:41.882 --> 01:10:43.202
The results are more or less random.
01:10:43.663 --> 01:10:48.946
We can also decide to go for another algorithm to check if this one would be more applicable to your use case.
01:10:54.488 --> 01:10:55.148
data sources.
01:10:55.268 --> 01:11:00.111
So data sources are actually the bread and butter of data scientists and machine learners.
01:11:01.172 --> 01:11:07.876
I don't think I have to go through every single one of them because they're also... And so the data sources, what point is he making here?
01:11:07.936 --> 01:11:19.543
Well, if Robert Malone says that the domain server used a computer-generated X-ray crystallography model of the 3CL protease and interfaced that model
01:11:20.543 --> 01:11:28.309
with all drugs and pharmaceuticals approved by the FDA and in the catalog, then the data is that list.
01:11:29.370 --> 01:11:33.853
And the data can't just be the list, right, or just the chemical symbols for it or the names.
01:11:35.034 --> 01:11:44.902
It's got to be a three-dimensional structure that would interact with the three-dimensional structure of the 3Cl protease that they simulated the X-ray crystallography model of.
01:11:46.084 --> 01:11:49.367
And somewhere in there, there's laser scanning confocal microscopy.
01:11:51.409 --> 01:12:01.619
But somehow or another, the data about all drugs and pharmaceuticals in the FDA catalog was fed into the domain server.
01:12:01.659 --> 01:12:02.940
And I'm very curious
01:12:03.948 --> 01:12:08.349
very, very curious as to the exact form of that data set.
01:12:08.389 --> 01:12:14.491
Because if we go back to his previous slides, there is data, but the data needs to be transformed.
01:12:14.531 --> 01:12:29.495
So if they got a list from the FDA, how did they transform that list into data, which could be usefully interfaced with the three-dimensional model of the 3Cl protease that they made in order to do this scan?
01:12:32.785 --> 01:12:52.412
I don't think you can really overestimate how much bullshit was contained in this little story about the domain server and the selection of Motadine, Silicoxib, Ivermectin, and Remdesivir.
01:12:55.069 --> 01:13:03.313
And as we hear more and more from this guy who's trying to outline how, it sounds really good, but it's not going to work as well as you imagine because of this.
01:13:04.834 --> 01:13:13.839
Because it's all basically human error that's going to pollute this because again, there's too many assumptions that are being inserted by humans.
01:13:13.899 --> 01:13:17.561
And now we're going to adapt it with more assumptions from humans.
01:13:17.601 --> 01:13:19.302
Well, maybe we should tweak this one a little bit.
01:13:21.055 --> 01:13:30.480
So how much understanding is happening and emerging from this process that would otherwise emerge from a model and an experiment and a model and an experiment?
01:13:32.382 --> 01:13:38.505
I see just a hyper-acceleration of the generation of useless knowledge.
01:13:40.566 --> 01:13:44.188
A hyper-acceleration of our disengagement with reality.
01:13:44.408 --> 01:13:45.009
That's what I see.
01:13:45.831 --> 01:14:02.765
And the idea that Robert Malone started his whole show by telling us a story about an AI program that had predicted that these things would be useful and that he was now advocating for their use, I mean, malevolence.
01:14:03.306 --> 01:14:04.086
Stop lying!
01:14:05.848 --> 01:14:07.289
It's completely underperforming.
01:14:07.469 --> 01:14:08.810
The results are more or less random.
01:14:09.291 --> 01:14:14.555
We can also decide to go for another algorithm to check if this one would be more applicable to your use case.
01:14:15.731 --> 01:14:16.411
data sources.
01:14:16.531 --> 01:14:21.373
So data sources are actually the bread and butter of data scientists and machine learners.
01:14:22.213 --> 01:14:26.874
Um, I don't think I have to go through every single one of them because they also address different needs.
01:14:27.714 --> 01:14:30.255
Um, he is German, right?
01:14:30.275 --> 01:14:41.879
So, I mean, it's going to be a certain level of, of meticulous because if you check, if you would like or mediocre or a target that is now of interest to you, you can go and check.
01:14:41.939 --> 01:14:43.199
Is there anything published already?
01:14:43.638 --> 01:14:51.983
and then just mine for anything that is active and also very, very important inactive to feature model with the data.
01:14:52.263 --> 01:15:02.609
So this is another thing, right, that I think can be emphasized here if I do it correctly, is that when you identified these compounds, you should have been able to verify it somehow.
01:15:02.650 --> 01:15:06.032
You should have made an argument about what their possible mechanism might be.
01:15:06.472 --> 01:15:09.634
So how to identify four disparate compounds that
01:15:10.134 --> 01:15:23.290
presumably have very different uses from their traditional perspective list on the FDA, now being repurposed to interact with the 3CL protease, he should have had some idea of what he would have expected them to do.
01:15:24.284 --> 01:15:26.226
Are they going to reduce the activity of this?
01:15:26.606 --> 01:15:28.247
Are they going to block the binding site?
01:15:28.587 --> 01:15:30.289
Are they going to disrupt its translation?
01:15:30.329 --> 01:15:32.430
What exactly did he think was going to happen?
01:15:32.471 --> 01:15:42.458
Because again, once you identify things, you have to kind of verify why it is that it would be and what, and, and, and decide whether or not the results of the AI are good or not.
01:15:42.498 --> 01:15:45.421
But yet somehow or another, it only took three weeks.
01:15:46.442 --> 01:15:50.645
And then they had this definitive list and Steve Kirsch was ready to raise money for them.
01:15:59.232 --> 01:16:03.195
On this, you can also take a screenshot or just wait for the YouTube upload.
01:16:04.015 --> 01:16:12.982
And there are also more databases obviously available that can be of use, uh, in regards to whatever you would like to achieve with your machine learning approach.
01:16:14.423 --> 01:16:19.266
I mean, we're really getting to the stage now where it feels like, Oh God, sorry.
01:16:20.868 --> 01:16:21.268
It's an ad.
01:16:25.882 --> 01:16:27.343
We're really getting to the stage now.
01:16:27.363 --> 01:16:30.346
So next, I would like to talk a little bit about the machine learning.
01:16:30.426 --> 01:16:30.947
Here we go.
01:16:31.627 --> 01:16:38.053
In ultimate theory, you could subdivide them into two parts that can be also combined a little bit.
01:16:38.874 --> 01:16:40.616
The first one is supervised learning.
01:16:40.836 --> 01:16:46.061
Supervised learning is kind of like using data that is labeled.
01:16:46.101 --> 01:16:48.843
This is the most important identifier for supervised learning.
01:16:49.484 --> 01:16:50.705
Labeled data means that you
01:16:51.716 --> 01:16:55.738
add a value on characterization of an entry.
01:16:55.938 --> 01:16:59.680
So you have molecule A, and you say molecule A is active.
01:17:00.501 --> 01:17:01.902
Molecule B, inactive.
01:17:02.002 --> 01:17:03.422
Molecule C, active.
01:17:03.663 --> 01:17:08.845
So supervised learning is kind of like setting parameters.
01:17:08.925 --> 01:17:15.989
Like you say, OK, I'm assuming that this reaction or this activity here is going to be linear.
01:17:19.676 --> 01:17:21.156
And so that's supervised learning.
01:17:21.176 --> 01:17:26.937
You're not letting the machine learning algorithm decide or come to the conclusion that it's linear.
01:17:27.357 --> 01:17:28.998
You're already setting it up that way.
01:17:29.998 --> 01:17:38.719
Another way to do it might be to say that these are active and these are not active, and teach them this classification a few times before it starts making that classification for itself.
01:17:39.200 --> 01:17:41.220
So that's, I think, is what he's saying here.
01:17:41.860 --> 01:17:47.061
E50 values, EC50 values, KI values, you name them, it can be a lot of things.
01:17:48.086 --> 01:17:50.847
What you can achieve with supervised learning is regression.
01:17:51.127 --> 01:17:57.410
So I think we can all recall things like the prediction of compound activity.
01:17:58.951 --> 01:18:05.234
Saying on the x-axis, for x1, you could have a principal component or even log p value.
01:18:05.754 --> 01:18:09.075
And on the left side, you could have another value like activity.
01:18:09.696 --> 01:18:14.638
So you could actually check for how does the activity of my compounds correlate
01:18:15.612 --> 01:18:17.994
between log p value and activity.
01:18:18.494 --> 01:18:25.799
And then use this to extrapolate for prediction of novel compounds in their activity.
01:18:26.440 --> 01:18:35.666
I think we all have been in the situation where we've been challenged with QSAR, so quantitative social activity relationship analysis.
01:18:36.346 --> 01:18:44.212
And this is one of those kinds of things where you just predict the activity of a social based on prediction.
01:18:45.575 --> 01:18:58.023
You could also use it to predict log p values, and machine learning algorithms have reached a status where their log p predictions are actually even better than the experimentally measured values.
01:18:58.424 --> 01:19:03.527
So this is where machine learning has actually reached a very, very satisfying status.
01:19:04.308 --> 01:19:07.590
On the other hand, you could also use for classification, which is the white figure.
01:19:08.410 --> 01:19:13.634
You could differentiate between active and inactive molecules, kind of like building clusters of compounds.
01:19:14.056 --> 01:19:17.019
You could also use this to predict toxic or non-toxic compounds.
01:19:19.221 --> 01:19:21.362
On the other hand, we have the unsupervised learning.
01:19:21.583 --> 01:19:23.384
Here, the input data is time-labeled.
01:19:23.424 --> 01:19:26.687
So basically, we don't actually know a lot about the data.
01:19:27.167 --> 01:19:30.250
Yet, we can use it to kind of like cluster compounds.
01:19:31.171 --> 01:19:33.473
One usage for that could be biological activity.
01:19:33.633 --> 01:19:38.057
So let's assume you would have a set of compounds with different parameters.
01:19:38.974 --> 01:19:51.617
And here we could say, well, the blue ones are dopamine receptor antagonists, the red ones are dopamine receptor agonists, and green ones are histamine receptor agonists based on the input data we provided it with.
01:19:52.377 --> 01:19:56.318
And it would then feed the whole system, the whole model with a new compound.
01:19:56.598 --> 01:20:01.639
We could actually deduct if this belongs to one of those two cases.
01:20:03.460 --> 01:20:06.761
Another idea would be to use clustering for molecule scaffolds.
01:20:07.281 --> 01:20:07.821
So basically,
01:20:08.597 --> 01:20:21.205
not only looking at the 2D structure, not only the kind of like obvious fact, but also to look a little bit, maybe a little bit into the topological side, like the volume and orientation in the room.
01:20:22.386 --> 01:20:24.707
Another nice thing is actually the association.
01:20:24.847 --> 01:20:31.271
So that you could associate a particular motive in a molecule to a biological activity.
01:20:31.371 --> 01:20:35.554
So can you correlate the presence of a pharmacophore
01:20:36.149 --> 01:20:38.950
to a desired effect of your compound.
01:20:40.131 --> 01:20:47.834
Furthermore, it's also something interesting that can be used for the prediction of synthesis route and adverse drug reactions.
01:20:48.194 --> 01:21:02.021
Maybe this is going too slow now, but the important thing to see here is that when somebody says they used an AI machine learning algorithm to identify four drugs to be repurposed at the beginning of the pandemic and you hear
01:21:03.120 --> 01:21:22.997
this talk you have to hear that he's a liar because remember he's supposed to also dress up and ride horses and raise emus and building barns and was just minding his own business at the foot of the Shenandoah National Forest when suddenly Michael Callahan called and said you better spin your team up
01:21:28.407 --> 01:21:33.308
Did he used unsupervised learning or supervised learning with regard to the domain server?
01:21:34.769 --> 01:21:36.509
It also has a little bit of background.
01:21:37.049 --> 01:21:40.290
I'm a little bit focused on the MedCamp perspective and small molecule design.
01:21:40.670 --> 01:21:48.372
Obviously, there are a lot of things that would be obviously in the higher field of drug discovery, kind of like phase one and phase two assessment of biological data.
01:21:48.732 --> 01:21:52.934
But this is not what this whole presentation is focused on.
01:21:55.994 --> 01:21:57.315
Are you a hot sleeper?
01:21:57.435 --> 01:21:57.595
No.
01:22:02.134 --> 01:22:03.275
They don't like me using this.
01:22:04.295 --> 01:22:06.016
Now, this is where the fun part starts.
01:22:06.096 --> 01:22:06.637
Here we go.
01:22:06.937 --> 01:22:08.778
A lot of people drop the ball.
01:22:09.938 --> 01:22:11.299
Let me dive into the deep learning.
01:22:11.979 --> 01:22:16.722
Deep learning is a little bit more abstract.
01:22:17.542 --> 01:22:24.946
The key feature of deep learning is that you enter high-dimensional data fields.
01:22:25.066 --> 01:22:27.667
Let's maybe start with the top example.
01:22:29.265 --> 01:22:30.986
convolutional neural networks.
01:22:31.166 --> 01:22:38.668
So convolutional, meaning coming, stemming from evolution and con, so evolving together.
01:22:39.608 --> 01:22:41.209
You take a picture, yeah?
01:22:41.229 --> 01:22:47.251
So convolutional neural... Shouldn't it be coevolution then instead of convolution?
01:22:47.391 --> 01:22:47.731
I don't know.
01:22:48.691 --> 01:22:50.132
Works extremely well for pictures.
01:22:51.212 --> 01:22:55.273
You take a picture and then you go through the picture and you tell what's the presence here?
01:22:55.293 --> 01:22:56.173
What's the color?
01:22:56.233 --> 01:22:57.034
What's the shape?
01:22:57.494 --> 01:22:58.094
What's the depth?
01:22:58.587 --> 01:23:01.008
of this particular area.
01:23:01.048 --> 01:23:11.930
They always have to go back to image processing because image processing is where they actually can kind of draw cartoons and kind of understand and also explain how the data will be broken up.
01:23:11.950 --> 01:23:13.570
You might take a black and white part.
01:23:13.590 --> 01:23:16.991
You might take all the blue, all the red, and then combine them and you know, whatever.
01:23:19.231 --> 01:23:20.311
What do you do with a protein?
01:23:22.572 --> 01:23:24.152
What do you do with a DNA sequence?
01:23:26.513 --> 01:23:27.633
What do you do with a proteome?
01:23:29.611 --> 01:23:30.912
What do you do with a genome?
01:23:31.012 --> 01:23:34.175
What do you do with a medical record from birth to death?
01:23:36.037 --> 01:23:42.543
How does that raw data get transformed into something useful, like an image is transformed into useful data?
01:23:44.125 --> 01:23:54.455
It's very easy for them to wave around about, well, we're great with, I mean, but all those, all those algorithms are based on KAPTCHA teaching.
01:23:56.266 --> 01:24:04.032
hundreds and hundreds of millions of people teaching the AI how to identify fire hydrants and wheels and buses and trains.
01:24:06.354 --> 01:24:07.395
That's not for nothing.
01:24:09.076 --> 01:24:18.103
That's data that they're feeding into a machine learning algorithm so that they can convince you, look, this stuff just makes great art.
01:24:19.844 --> 01:24:23.247
It's still just bullshit image processing and
01:24:25.055 --> 01:24:26.116
and Brute Force.
01:24:27.917 --> 01:24:34.300
Possibly also a room full of people on Photoshop competing for the first three examples.
01:24:38.282 --> 01:24:49.028
Man, oh man, the Domain Server, the Domain Program, Robert Malone, David Hone, Steve Kirsch, Meryl Nass,
01:24:50.629 --> 01:24:55.010
The diffuse proposal, it's all part of the same show.
01:24:55.810 --> 01:24:59.171
Elaborate theater, Lollapalooza of liars.
01:25:00.072 --> 01:25:19.737
And if you understand anything about the limitations of AI and machine learning and anything about how it's done, the domain server, the domain program, the DITRA-RAN that was published in multiple places is one of the first pieces of the
01:25:21.561 --> 01:25:25.322
pandemic narrative curation team?
01:25:28.483 --> 01:25:36.686
For each section and each area of the image, you construct one of those little representative numerical values.
01:25:37.546 --> 01:25:40.247
Those convolutional layers are then stapled.
01:25:40.807 --> 01:25:46.429
And then what the idea behind this algorithm is, well, are they all really necessary?
01:25:46.989 --> 01:25:49.610
So you actually reduce the layers here
01:25:50.083 --> 01:25:51.143
So a smaller number.
01:25:51.343 --> 01:26:00.465
So you generate a pool of features that still capture the whole idea of the image, but are smaller, include less dimensions.
01:26:01.405 --> 01:26:02.185
And then you keep on going.
01:26:02.205 --> 01:26:08.707
And so we're going to take an irreducible complexity and reduce it to an essential number of dimensions.
01:26:08.787 --> 01:26:15.528
And then we're going to claim that using those essential number of dimensions, we're going to have insight into the irreducible complexity.
01:26:18.453 --> 01:26:26.456
You look again at this pool, and then reduce it to another convolutional layer, and then check, well, what can we remove?
01:26:26.896 --> 01:26:27.837
And so on and so forth.
01:26:27.877 --> 01:26:29.878
And you do it a lot of steps, a lot of times.
01:26:31.398 --> 01:26:34.039
Finally, you end up with a final pool.
01:26:35.160 --> 01:26:35.860
And what you do then?
01:26:36.120 --> 01:26:36.960
You flatten the pool.
01:26:37.060 --> 01:26:43.303
So basically, you try to reduce this multidimensional layer into a one-dimensional
01:26:44.752 --> 01:26:53.657
So only one dimension, meaning you will only include only numbers, a sequence of numbers, long, long, long numbers, but a sequence of numbers nevertheless.
01:26:56.038 --> 01:27:01.922
Finally, you go into this fully connected layer that can actually read this flattened layer.
01:27:02.222 --> 01:27:09.286
So this layer is based on information, the data that you use to train your model, and what it does then, it reads this flattened layer.
01:27:09.366 --> 01:27:13.629
It then compares, what are the patterns of this flattened layer that I can actually
01:27:14.264 --> 01:27:15.164
read and understand.
01:27:16.605 --> 01:27:17.205
What happens then?
01:27:17.526 --> 01:27:18.446
You receive an output.
01:27:18.726 --> 01:27:21.187
So it will tell you, this is a picture of a cute dog.
01:27:22.068 --> 01:27:24.489
I mean, like maybe some of you have seen it.
01:27:25.790 --> 01:27:32.213
You can train actually one of those CNNs to recognize, to differentiate between chihuahuas and a blueberry muffin.
01:27:32.973 --> 01:27:34.714
Yeah, so that's kind of like the most
01:27:35.933 --> 01:27:36.934
prominent use case.
01:27:37.114 --> 01:27:41.837
And this is definitely proof that we can use AI to search for drunks.
01:27:42.097 --> 01:27:51.223
And that's why Robert Malone did it in three weeks with a team of volunteers and identified remdesivir, famotidine, silicoxib, and ivermectin.
01:27:51.723 --> 01:27:52.424
Stop lying!
01:27:54.285 --> 01:27:55.186
So far at least.
01:27:56.947 --> 01:27:57.047
Um,
01:27:58.260 --> 01:28:03.022
Another interesting application is actually the deep learning neural network.
01:28:03.643 --> 01:28:09.966
Again, you start with an input layer and then you construct a neural network.
01:28:09.986 --> 01:28:12.147
So everything is connected with each other.
01:28:12.967 --> 01:28:23.272
What happens here, you extract different informations, different parameters from each input layer and then feed one of those neurons, basically one of those hidden layers.
01:28:23.892 --> 01:28:25.753
And then they feed each other again and again
01:28:26.280 --> 01:28:29.941
where you transmit one information from one neuron to another.
01:28:30.902 --> 01:28:41.305
So, what happens here, again, we have a lot of layers, we have a lot of dimensions going on, but by the end of the day, we again reduce them to a simple representation.
01:28:41.665 --> 01:28:46.567
And this representation understands a lot of connections happening in between.
01:28:47.047 --> 01:28:53.469
So, chatGPT is actually a deep learning neural network that use this kind of approach.
01:28:54.922 --> 01:29:09.793
And so again, dropping in the hint that language, deep language, these deep learning networks, these language networks are evidence that this technology, if we have enough computers, can be applied to irreducible complexity.
01:29:11.354 --> 01:29:24.378
Please understand that language is not an irreducible complexity, and the difference between muffins and a chihuahua in a two-dimensional photograph is not an irreducible complexity.
01:29:24.438 --> 01:29:26.578
These are very tractable problems.
01:29:27.139 --> 01:29:30.680
Language has been a puzzle that was cracked already in the 20s.
01:29:31.160 --> 01:29:34.761
That's why Edward Bernays was so succinct when he said,
01:29:35.481 --> 01:29:42.025
that the conscious and intelligent manipulation of the organized opinions and habits of the masses is an essential part of democratic governance.
01:29:45.768 --> 01:29:56.915
They are telling us stories, ladies and gentlemen, and building mythologies where we don't even need the scientific method anymore, even though we haven't used it correctly for 25 or 30 years in most domains.
01:29:57.936 --> 01:29:59.137
We don't even need it anymore.
01:30:00.310 --> 01:30:07.152
We haven't been applying it correctly for 25 or 30 years, but we don't even need it anymore because we've got AI now.
01:30:07.812 --> 01:30:08.613
Stop lying.
01:30:10.653 --> 01:30:25.438
You can do, and what this whole deep learning approach can achieve, it can understand a lot of different connections and a lot of different references between different data points.
01:30:25.936 --> 01:30:33.998
So ultimately it can connect Apple to the taste of sweet without having both of them standing next to each other in a long, long text.
01:30:34.718 --> 01:30:38.799
Just mining from a lot of different sources, a lot of different inputs.
01:30:41.680 --> 01:30:47.161
And what this can achieve is actually it can also detect a lot of things that you didn't explicitly request for.
01:30:47.342 --> 01:30:50.522
We see that a lot in our subsequent examples.
01:30:52.803 --> 01:30:52.903
So,
01:30:53.482 --> 01:30:56.464
Now we have a basic idea about machine learning in itself.
01:30:56.744 --> 01:30:58.405
A lot of data, a lot of processing.
01:30:59.246 --> 01:31:00.507
Now let's go back to pharma.
01:31:00.727 --> 01:31:06.831
So the thing with pharma is we are extremely, extremely bad.
01:31:07.452 --> 01:31:12.395
No other industry accepts such high failure rates as we do.
01:31:13.016 --> 01:31:17.879
So basically, drug discovery has an abysmal failure rate of approximately 90%.
01:31:18.099 --> 01:31:21.842
And this is based purely on phase one clinical trials.
01:31:22.734 --> 01:31:29.797
You can also see that 97% of cancer drugs actually fail during... 97% of cancer drugs fail.
01:31:29.837 --> 01:31:31.958
So what happens to all the people in those trials?
01:31:37.340 --> 01:31:41.641
Oh, you mean like the adenovirus treatment that they did for Jesse Gelsinger wasn't a one-off?
01:31:41.661 --> 01:31:43.862
You mean they're doing it to cancer people all the time?
01:31:44.102 --> 01:31:44.743
Oh, wow!
01:31:44.823 --> 01:31:45.663
I never knew!
01:31:45.683 --> 01:31:48.144
Yeah, right.
01:31:48.464 --> 01:31:50.005
It was because they are not able to meet endpoints.
01:31:51.417 --> 01:31:53.239
And there are a lot of things that contribute to that.
01:31:53.439 --> 01:32:03.349
One of those is obviously the lack of clinical efficiency, basically because the drugs do whatever they do, so they inhibit the enzyme, but
01:32:03.920 --> 01:32:10.748
once you apply them on a broader scope with a lot more layers, a lot more complexity because the human body is quite complex.
01:32:11.368 --> 01:32:13.871
Oh, is the human body quite complex?
01:32:14.011 --> 01:32:24.583
Do you mean to tell me that if the domain server identified ivermectin, for example, it might be not the right thing because although it interacts with the enzyme in the model, it also kills ants.
01:32:25.816 --> 01:32:30.758
It can also like make people, I don't know, sterile or something at the doses that they're suggesting.
01:32:31.138 --> 01:32:44.704
And so maybe it wasn't so good for Brett Weinstein to say that he was pretty sure that prophylactic ivermectin was the final way to get out of this and that zero COVID was still doable when Robert Malone came on his stream in June of 2021.
01:32:50.137 --> 01:32:58.920
So the unmanageable toxicity of ivermectin is being completely ignored, even though somehow or another it can cure cancer and also save healthy people.
01:33:01.881 --> 01:33:04.742
How does it work for cancer, but not hurt healthy people?
01:33:05.122 --> 01:33:09.443
I don't know very many cancer drugs that are good for people in general.
01:33:11.624 --> 01:33:13.825
You will end up with a lot of, yeah.
01:33:15.683 --> 01:33:18.425
Surprisingly, toxicity becomes also a problem.
01:33:19.366 --> 01:33:34.860
I think this is relating to the phase one clinical trials per se, because people do not expect side effects because off-targets were not realized back then, or the dosage is not as, as was extrapolated from the models.
01:33:35.820 --> 01:33:42.506
Surprisingly, which is a good thing, the poor drug-like properties are only amounting to 10 to 15%,
01:33:44.151 --> 01:33:48.814
based on also a lot of things that happened in the preclinical trials and the preclinical optimization.
01:33:48.974 --> 01:33:50.575
Oh, I assure you that's what I'm saying.
01:33:50.615 --> 01:34:03.681
They've been testing drugs on cancer people since the dawn of cancer, really, since the special cancer virus program, since Kaprowski and Plotkin and Baltimore and Gallo started this whole mess.
01:34:05.923 --> 01:34:10.145
What also surprised me a little bit is actually the lack of commercial needs and the strategic planning, but
01:34:10.526 --> 01:34:16.009
This is not something we as scientists and the Dr. Scholar field can influence too much.
01:34:17.669 --> 01:34:28.274
What I found pretty interesting is actually that the whole research and development procedure has become less efficient than the past.
01:34:29.014 --> 01:34:34.557
In fact, that's the thing that... It's becoming worse, you see.
01:34:35.037 --> 01:34:36.198
...has been researched already.
01:34:37.461 --> 01:34:42.464
If you think about that, we have all of those huge high throughput screening libraries and so on.
01:34:43.225 --> 01:34:48.729
And most of the easy compounds that are easy to synthesize have been also tested on the easy targets.
01:34:49.449 --> 01:34:59.316
So now, if you still would like to market a drug or get approval for a drug, you'd need to put more effort than you had to do in the previous decades.
01:35:00.817 --> 01:35:07.241
So ultimately, we see that if we could actually improve the prediction to the clinical efficiency,
01:35:07.679 --> 01:35:15.506
toxicity, and so on and so forth, we could actually save a lot of money and also accelerate the development of future drug candidates.
01:35:15.686 --> 01:35:19.790
In this little figure and graph, you can actually see how much time we can save.
01:35:20.210 --> 01:35:21.611
This is an assessment.
01:35:22.212 --> 01:35:26.275
And on the left side, we actually see the millions that you can save.
01:35:26.355 --> 01:35:28.717
So here, 500 millions that you could save.
01:35:29.598 --> 01:35:31.059
And here are the colors.
01:35:31.079 --> 01:35:31.620
So basically,
01:35:32.099 --> 01:35:33.079
And the early stages?
01:35:33.480 --> 01:35:37.541
I agree wholeheartedly that it is embarrassing that this is considered knowledge.
01:35:37.601 --> 01:35:38.702
There's no knowledge here.
01:35:38.742 --> 01:35:39.942
This is a mythology.
01:35:40.062 --> 01:35:45.464
He's only reiterating what he understands is his best take on that mythology.
01:35:45.765 --> 01:35:56.529
And he is extending it now for them by saying that there's this problem of our research and development in pharmaceutical companies doesn't work very well because the low-hanging fruit have long been taken.
01:35:59.161 --> 01:36:05.366
So we're going to have to work harder to reduce these costs that are more or less here.
01:36:05.707 --> 01:36:11.291
We can actually see that the speed is not a problem, so we won't save too much money there.
01:36:11.772 --> 01:36:15.855
But we actually need to improve the quality of the predictions.
01:36:16.316 --> 01:36:17.977
And this is especially true for preclinical.
01:36:17.997 --> 01:36:19.899
It becomes even more important in clinical phase two.
01:36:20.899 --> 01:36:23.782
And actually, if we are able to reduce the costs.
01:36:24.215 --> 01:36:33.469
And cost could be actually the acquisition of compounds, including the purchase and also the synthesis, but could only quality becomes a problem.
01:36:33.790 --> 01:36:36.734
And if we could increase the failure, reduce the failure rate here by 20%,
01:36:37.790 --> 01:36:41.231
we would actually save up close to 450 million.
01:36:41.291 --> 01:37:03.157
It's interesting to note that the company that Sasha sold in order to become retired and be able to focus on parenting her kids that were on the internet before the pandemic and featured on Alex Jones, she actually sold a company to Pfizer that was kind of about optimizing this, right?
01:37:03.988 --> 01:37:10.011
I mean, optimizing clinical trials and the effectiveness or the, the, oh, a little tiny.
01:37:10.031 --> 01:37:13.393
Oh, that's the smallest chipmunk baby I've ever seen.
01:37:13.513 --> 01:37:14.374
Holy cow.
01:37:16.435 --> 01:37:19.757
That's, it's like a micro chipmunk.
01:37:21.277 --> 01:37:24.519
I've never seen anything more cute than that out my back door in my life.
01:37:24.819 --> 01:37:25.520
Oh my goodness.
01:37:27.581 --> 01:37:29.022
Wow, that is hilarious.
01:37:29.222 --> 01:37:33.364
It's like, it was like a third the size of the chipmunks I see around here, maybe even a quarter.
01:37:34.107 --> 01:37:35.148
It was a micro chipmunk.
01:37:36.970 --> 01:37:39.752
US dollars per drug approval.
01:37:40.633 --> 01:37:50.402
And this is actually where I would assume most companies are now aware of, because this is where the target selection is actually the problem.
01:37:50.862 --> 01:37:58.389
So a lot of pharmaceutical companies have now restructured how they do the target validation steps here.
01:37:59.233 --> 01:38:05.237
And they now put a little more effort to select which targets should be pursued for a particular treatment.
01:38:05.617 --> 01:38:12.101
So this is actually already an implementation in the pharmaceutical companies around the globe.
01:38:13.962 --> 01:38:17.243
So now to the AI in drug discovery.
01:38:17.283 --> 01:38:19.685
So the first coming out of the drug discovery schema.
01:38:20.045 --> 01:38:23.207
And I think in the past few years... Notice in the middle there, NVIDIA.
01:38:24.804 --> 01:38:32.030
NVIDIA remember that that Nancy Pelosi and a bunch of other chumps made a lot of money on NVIDIA at the beginning of the pandemic why?
01:38:33.030 --> 01:38:40.996
because of processing power Not because people were playing a lot of games and and and a new Call of Duty came out.
01:38:41.077 --> 01:38:45.660
It's because They're using it in AI and drug discovery.
01:38:45.700 --> 01:38:51.384
You see do you see I see All have been a LinkedIn or other
01:38:52.378 --> 01:38:55.179
social media platforms and whatever it may be.
01:38:55.939 --> 01:39:01.880
And you can actually see how people are celebrating AI to solve every single problem that is out there.
01:39:03.600 --> 01:39:15.863
I mean, even if the financial newspapers and journals jump up on that, you can actually feel that there is a lot of money flowing in this direction.
01:39:16.703 --> 01:39:17.603
And it's actually true.
01:39:17.743 --> 01:39:20.184
So by 2021, approximately 16 billion have been spent
01:39:23.053 --> 01:39:29.016
on AI and drug discovery relating to collaborations.
01:39:30.276 --> 01:39:37.360
So are we seeing something ramping up that we're supposed to believe is AI drug discovery when it's actually not?
01:39:38.240 --> 01:39:39.621
Because again, think about this.
01:39:40.041 --> 01:39:41.502
They already have all the data.
01:39:42.322 --> 01:39:48.705
Robert Malone said they had a whole catalog from the FDA of all the approved drugs and nutraceuticals.
01:39:50.051 --> 01:39:51.232
So they have that data.
01:39:51.932 --> 01:39:59.897
They don't need any more of that data unless they are identifying novel compounds that they could synthesize, that they could create.
01:40:01.738 --> 01:40:12.044
In other words, what I'm trying to understand is, is if they're investing all of this money and all this computing power, when all of the low-hanging fruit is gone,
01:40:15.151 --> 01:40:26.577
Why would they invest $16 billion unless they were collecting more data than just the FDA chemical list or a x-ray crystallography of a viral protein?
01:40:28.338 --> 01:40:33.802
What data would they be feeding into these AIs in order to do drug discovery?
01:40:38.184 --> 01:40:43.247
They're feeding in human genome and proteome data.
01:40:45.646 --> 01:40:47.487
Probably gathered by the swabs.
01:40:48.628 --> 01:40:54.752
Probably gathered by all the remnants that are produced in every hospital all the time, all the time, all the time.
01:40:57.014 --> 01:40:58.915
Because domain already exists.
01:40:58.975 --> 01:41:00.836
Domain can find drugs.
01:41:00.976 --> 01:41:01.757
We have that.
01:41:01.837 --> 01:41:03.338
It works in like three weeks.
01:41:03.418 --> 01:41:06.560
We have the data set that they use from the FDA.
01:41:06.600 --> 01:41:10.323
We don't need to make new data sets for all the chemicals and stuff, do we?
01:41:12.014 --> 01:41:15.397
So what data are they feeding into these AIs?
01:41:15.457 --> 01:41:16.678
What the data would be?
01:41:16.778 --> 01:41:17.618
What would it be?
01:41:17.638 --> 01:41:25.364
It would be most definitely human data from the human experiments that are being run right now.
01:41:27.566 --> 01:41:35.232
And while we argue about viruses and whether or not Nick Hudson was lying about Pfizer data,
01:41:37.557 --> 01:41:42.899
We aren't seeing what's actually happening right in front of us, was shown to us before the pandemic.
01:41:42.939 --> 01:41:47.261
They are taking our data and they're using it already now.
01:41:47.341 --> 01:41:49.122
It's not gonna start.
01:41:50.823 --> 01:42:00.567
It started with the college kids and the daily and weekly swabs that were sold by the University of Illinois and the University of Ohio and all these other places.
01:42:01.987 --> 01:42:05.249
Make no mistake about it, ladies and gentlemen, we are at the stage now
01:42:06.826 --> 01:42:15.892
where we can tell the truth and it's unstoppable.
01:42:16.892 --> 01:42:20.715
Acquisitions of companies, and those are only the reported deals.
01:42:20.775 --> 01:42:23.497
So the true number will be much, much, much higher.
01:42:24.317 --> 01:42:33.643
But if you think about that, that one FDA approved drug that costs approximately 2.8 billion US dollars, you could actually ask yourself, well,
01:42:34.372 --> 01:42:40.636
We have invested 16 billion by 2021 and even more in 2023.
01:42:41.436 --> 01:42:46.079
So where are my FDA approved drugs that have been discovered by AI approaches?
01:42:46.699 --> 01:42:47.240
Where are they?
01:42:47.480 --> 01:42:49.961
And yeah, it's a thing.
01:42:50.542 --> 01:42:57.106
I mean, it took the domain server four weeks to identify four drugs that interact with a new novel virus protein.
01:42:58.365 --> 01:43:05.027
So why in God's green earth can't we use domain or a server like it or a AI like it to solve a lot of these problems?
01:43:05.727 --> 01:43:07.087
We have all these targets.
01:43:07.747 --> 01:43:13.249
Robert Malone knows how to make an X-ray crystallography of a model of any protein, so why not?
01:43:15.449 --> 01:43:23.811
They spent $16 billion already since in 2021 to do it on their own when AIs and domain already could do it in three weeks?
01:43:24.752 --> 01:43:25.852
With a volunteer team?
01:43:27.017 --> 01:43:28.358
Run by an emu breeder?
01:43:32.302 --> 01:43:34.284
Ladies and gentlemen, I don't think we need to go any farther.
01:43:34.304 --> 01:43:38.468
I think this is a good place to take a pause so that I can get to basketball.
01:43:39.709 --> 01:43:41.451
I do think it's almost time to leave.
01:43:41.511 --> 01:43:43.713
I think my alarm was about to go off.
01:43:45.395 --> 01:43:49.579
Ladies and gentlemen, I would strongly make the argument that
01:43:55.595 --> 01:44:06.217
I would strongly make the argument that there is an illusion here, and it's much more elaborate than any of us were aware of at the beginning of the pandemic, and that illusion is being sustained by people that are essentially traitors.
01:44:06.257 --> 01:44:21.921
They are selling our kids and our country to these globalist slavers, and I think we're starting to really be able to poke at them hard, and that's why what happened in the previous broadcast happens, because they have very little left.
01:44:23.300 --> 01:44:36.188
other than to try and create rivalries that other people pay attention to, create enemies and wars that don't matter about ideas that don't matter.
01:44:37.831 --> 01:44:39.833
that's why that French guy came on here.
01:44:40.334 --> 01:44:45.340
Maybe he meant well because he really felt attacked, but he wasn't attacked and I wasn't attacking him.
01:44:45.360 --> 01:44:58.054
I was attacking all of the anonymous accounts that use mice in their logo because the vast majority of them are indeed meddlers, which he actually admitted but then insisted on coming back to the
01:44:58.835 --> 01:45:14.105
Nick Hudson's a liar and Nick Hudson lied about Pfizer and Pfizer can be put away if we just realize that Sasha's bad or something like that and make basically making us talk about something else other than the fundamental biology that we've been teaching for four years.
01:45:14.426 --> 01:45:17.508
He made very clear that he didn't want to discuss any biology with me.
01:45:17.868 --> 01:45:24.592
He wanted to make very clear that I apparently attacked him and that's why he could put a crack pipe in my mouth on Twitter.
01:45:25.013 --> 01:45:28.335
This is the level of discourse that has left for them.
01:45:29.256 --> 01:45:33.839
They can't put Robert Malone out there and have him discredit my take on clones.
01:45:34.220 --> 01:45:39.424
They can't have Jessica Rose go out there and discredit my take on transfection.
01:45:39.824 --> 01:45:49.893
And they can't have Bret Weinstein come out there and explain on his podcast succinctly why RNA can indeed pandemic if it has a fear and cleavage site or a few HIV inserts.
01:45:50.793 --> 01:45:52.835
They can't because they are liars.
01:45:52.895 --> 01:45:53.536
Stop lying!
01:45:56.041 --> 01:46:02.247
Ladies and gentlemen, intramuscular injection of any combination of substances with the intent of augmenting the immune system is dumb.
01:46:02.328 --> 01:46:07.293
Transfection in healthy humans is criminally negligent and RNA cannot pandemic.
01:46:07.313 --> 01:46:14.100
So please stop all transfections in humans because they are trying to eliminate the control group by any means necessary.
01:46:14.580 --> 01:46:19.922
If you like what you saw today, please find a way to share this from GigaOMBiological.com.
01:46:20.382 --> 01:46:26.405
GigaOM.bio has a certificate problem right now, so your browser will probably warn you about it.
01:46:27.105 --> 01:46:29.106
But that's okay, we'll fix that in a day or two.
01:46:29.626 --> 01:46:36.429
And Stream.GigaOM.bio is where you find all the replays, complete with the music, no censorship, no sign-up necessary.
01:46:37.029 --> 01:46:42.651
easy to share, and the links that you share on Twitter will show up in Twitter, which is what Twitter likes.
01:46:43.051 --> 01:46:49.353
So if you want to share the stream, one way to do it is to grab that link in stream.gigaom.bio and drop it into Twitter.
01:46:49.693 --> 01:46:50.713
I love you all very much.
01:46:50.793 --> 01:46:53.134
Thanks for tolerating my earlier performance.
01:46:53.174 --> 01:46:56.655
I know it wasn't perfect, but at least I'm
01:46:57.515 --> 01:46:59.117
I'm just me.
01:46:59.417 --> 01:47:00.138
I'm just Jay.
01:47:00.518 --> 01:47:03.662
And you know exactly what you're going to get when you come here if you've been here for a while.
01:47:03.702 --> 01:47:05.503
So thank you very much to all my supporters.
01:47:06.585 --> 01:47:10.328
The people that are subscribing on Substack, please understand that I'm going to get there.
01:47:10.789 --> 01:47:15.414
It's just really hard to do all of this stuff that these other people seem to do very easily.
01:47:15.914 --> 01:47:20.418
Lots of pages on Substack, lots of tweets all day, lots of podcasts.
01:47:20.498 --> 01:47:31.986
It's all done with a staff and this is just me and I've got my best friend in the background taking care of the kids and making sure that the food's on the table and bills are paid and I just, I get to be this biologist.
01:47:32.547 --> 01:47:40.853
So if anybody is in need of a thank you, it's my wife because without her unwavering belief,
01:47:43.951 --> 01:47:48.655
I would be pretty sad, so I don't know what to say other than thanks, Fearland.
01:47:48.695 --> 01:47:49.996
Thanks, everybody, for supporting me.
01:47:50.016 --> 01:47:51.077
I got basketball to do.
01:47:51.537 --> 01:47:51.938
See you guys.