WEBVTT - Faster, Cheaper Drugs with AI 0:00:15.316 --> 0:00:23.036 Pushkin. Over the past few decades, it's become more and 0:00:23.156 --> 0:00:27.076 more expensive to develop new drugs. It now costs over 0:00:27.116 --> 0:00:30.276 a billion dollars on average to bring a new drug 0:00:30.316 --> 0:00:33.316 to market in the United States, and of course drug 0:00:33.356 --> 0:00:36.796 companies pass those high development costs onto us in the 0:00:36.836 --> 0:00:40.076 form of higher drug prices. This has been going on 0:00:40.116 --> 0:00:42.156 for so long that we have sort of gotten used 0:00:42.196 --> 0:00:46.916 to it. But when you zoom out, it's strange because, 0:00:47.196 --> 0:00:49.276 as I've said before on this show, and as I 0:00:49.316 --> 0:00:52.076 will say again on this show, one of the main 0:00:52.156 --> 0:00:57.276 things technology does is it makes things more efficient and 0:00:57.356 --> 0:01:01.916 therefore cheaper. Over the past few centuries, we've seen technologies 0:01:01.956 --> 0:01:04.996 make all kinds of things cheaper, everything from clothes to 0:01:05.116 --> 0:01:10.516 food to TVs. So why hasn't new technology made drugs cheaper? Two. 0:01:15.116 --> 0:01:17.756 I'm Jacob Goldstein and this is What's Your Problem, the 0:01:17.796 --> 0:01:20.276 show where I talk to people who are trying to 0:01:20.316 --> 0:01:24.836 make technological progress. My guest today is Alice Zang, co 0:01:24.996 --> 0:01:29.316 founder and CEO of verge Genomics. Alice's problem is this, 0:01:30.036 --> 0:01:33.676 how do you use artificial intelligence to drive down the 0:01:33.756 --> 0:01:38.636 price of discovering and developing new drugs? Why is it 0:01:38.676 --> 0:01:41.196 getting more expensive to develop drugs, despite the fact that 0:01:41.236 --> 0:01:44.396 we have better technology to do it. Yeah. Absolutely. One 0:01:44.436 --> 0:01:46.996 of the reasons is, you know, even though a lot 0:01:47.036 --> 0:01:51.276 of the new technologies we've developed have made us better 0:01:51.596 --> 0:01:56.316 at testing more drugs faster, but the fundamental problem is 0:01:56.356 --> 0:01:58.236 that even if we can get a drug all the 0:01:58.276 --> 0:02:00.756 way to clinical trials, which is the last step of 0:02:00.836 --> 0:02:04.916 drug development, ninety percent of those drugs still fail. So 0:02:04.956 --> 0:02:07.396 if you think about it, we're spending millions on each drug. 0:02:09.236 --> 0:02:11.636 Of those drugs are failing at the last and most 0:02:11.676 --> 0:02:15.276 expensive stage of drug development. And so really most of 0:02:15.396 --> 0:02:18.996 that billion plus dollar figure you hear is due to 0:02:19.036 --> 0:02:22.676 the cost of failure. Just to be clear, that figure 0:02:22.716 --> 0:02:25.236 more than a billion dollars. It's you've got to include 0:02:25.276 --> 0:02:28.036 the cost of all the drugs that don't work exactly, 0:02:28.196 --> 0:02:30.396 the ones that do right exactly. So the ones that 0:02:30.436 --> 0:02:32.356 do work have to pay for all the ones that fail. 0:02:32.476 --> 0:02:36.396 That's the fundamental problem, exactly, And you're setting out to 0:02:36.996 --> 0:02:40.836 fix that if you can. Absolutely, we think there's an 0:02:40.876 --> 0:02:47.156 opportunity for AI to fundamentally shift really the failure rate, 0:02:47.716 --> 0:02:50.116 and the most impactful time to do that really is 0:02:50.116 --> 0:02:54.436 the failure in clinical trials. So can we predict before 0:02:54.476 --> 0:02:58.996 we go in to these expensive clinical trials genes or 0:02:59.116 --> 0:03:02.916 targets or drugs that are more likely to work in humans, 0:03:02.956 --> 0:03:07.076 because even a ten percent decrease in that failure rate 0:03:07.116 --> 0:03:11.076 could have massive I saw a number of up to 0:03:11.116 --> 0:03:15.636 fifteen billion dollars annually in industry cost savings. You could 0:03:15.636 --> 0:03:18.196 still be in a universe where most of the drugs 0:03:18.236 --> 0:03:20.756 that go into clinical trials fail, but instead of ninety 0:03:20.796 --> 0:03:23.636 percent of them failing, seventy percent of them fail, and 0:03:23.676 --> 0:03:25.356 that would be a huge win. That would be a 0:03:25.436 --> 0:03:28.476 huge efficiency gain. It would save a ton of money, absolutely, 0:03:28.476 --> 0:03:31.636 And I think that's something that's underappreciated about AI and 0:03:31.756 --> 0:03:35.756 really any technology, is that oftentimes people have this expectation 0:03:36.236 --> 0:03:40.156 that this technology is going to absolutely transform a field overnight. 0:03:40.476 --> 0:03:42.716 And I think what people don't appreciate is that most 0:03:42.716 --> 0:03:45.396 of the time that doesn't happen. It's always step by 0:03:45.436 --> 0:03:49.836 step incremental. But even a ten percent change would have 0:03:49.876 --> 0:03:52.876 billions of dollars of cost savings and would be a 0:03:52.956 --> 0:03:56.156 huge win for patients in the industry worldwide. I like 0:03:56.316 --> 0:03:59.596 that frame, actually, I like that frame of maybe AI 0:03:59.756 --> 0:04:03.836 can have drugs fail most of the time, but not 0:04:03.916 --> 0:04:05.956 as much of the time as they fail. Now, like, 0:04:05.996 --> 0:04:08.396 it seems very credible, It seems very plausible. Would you 0:04:08.396 --> 0:04:10.836 put it that way? Yeah, it's all life is nothing 0:04:10.876 --> 0:04:15.596 but a learning process, Yes, getting less bad at everything. 0:04:16.756 --> 0:04:20.356 So I know you were studying to be a doctor 0:04:20.396 --> 0:04:22.876 and a researcher not that long ago, a few years 0:04:22.876 --> 0:04:25.316 ago before you started your company. Like, tell me how 0:04:25.356 --> 0:04:30.636 you went from an mdphd program to starting the company. Well, 0:04:30.716 --> 0:04:35.956 my PhD research was actually in using genomic analysis and 0:04:36.076 --> 0:04:41.476 computational biology to analyze large scale data sets and find 0:04:41.516 --> 0:04:46.516 new drugs that could improve drug development. And we found 0:04:46.556 --> 0:04:50.836 that from our very first drug that was predicted from 0:04:50.836 --> 0:04:53.956 our algorithms when we put it in mice after they've 0:04:53.996 --> 0:04:58.316 been injured, help them walk and recover from that injury, 0:04:58.476 --> 0:05:01.396 that nerve injury about four times faster than the leading standard. 0:05:01.436 --> 0:05:03.796 And that was just the first drug that was predicted. 0:05:03.876 --> 0:05:06.276 And I looked at this technology in this approach and 0:05:06.316 --> 0:05:09.276 I thought, Wow, there's so much promise here. You know, 0:05:09.316 --> 0:05:12.316 am I really going to just publish this and let 0:05:12.316 --> 0:05:16.436 it sit on a bookshelf somewhere, or if I'm not 0:05:16.476 --> 0:05:19.116 going to be the one to really develop this to patients, 0:05:19.196 --> 0:05:21.636 you know who will, And when I looked out off 0:05:21.636 --> 0:05:25.196 the field, I did not see a ton of biotech 0:05:25.276 --> 0:05:29.436 or farmer companies that were truly computationally driven. Usually within 0:05:29.476 --> 0:05:33.356 pharma companies they might bring in computational biologists to support. 0:05:34.116 --> 0:05:38.036 There are scientists or their biologists, but there wasn't really 0:05:38.156 --> 0:05:41.596 a genomics computationally driven company at that time. Now there 0:05:41.636 --> 0:05:44.476 are many, but at the time there are very few. 0:05:44.956 --> 0:05:47.676 And so I actually, you know, it wasn't a binary decision. 0:05:47.716 --> 0:05:50.596 People always ask me, how did you make the courageous 0:05:50.596 --> 0:05:55.556 decision to leap? It wasn't really like that. I think 0:05:55.596 --> 0:05:57.916 what we did first is that we just took three 0:05:57.956 --> 0:06:00.596 months three month leave of absence. We joined a program, 0:06:00.636 --> 0:06:04.036 an incubator called a y combinator. We as you and 0:06:04.396 --> 0:06:08.916 you and well me and my co founder Jason. And 0:06:09.316 --> 0:06:11.996 the first question really was, you know, can we even 0:06:13.156 --> 0:06:17.636 generate some data that validates that computational biology can predict 0:06:17.676 --> 0:06:21.036 targets that work? And then when we saw some data, 0:06:21.116 --> 0:06:25.116 the next question was can we even hire people that 0:06:25.236 --> 0:06:27.436 want to come on? And the next question was can 0:06:27.516 --> 0:06:31.316 we even raise money from people that will care? And 0:06:31.356 --> 0:06:34.236 I think that is so such an important lesson because 0:06:34.276 --> 0:06:37.916 I think people oftentimes get caught up in just the destination, 0:06:38.036 --> 0:06:39.676 you know, is where I want to be? Is this 0:06:39.756 --> 0:06:42.956 the career I want to have that they don't take 0:06:42.996 --> 0:06:45.156 the first step, And really it's the first step that's 0:06:45.196 --> 0:06:48.036 needed to actually get the data to even decide if 0:06:48.076 --> 0:06:50.916 it's the appropriate track for you. And did you really 0:06:50.956 --> 0:06:53.116 just keep thinking, well, this might not work, but let's 0:06:53.196 --> 0:06:55.156 do the next thing. Were you in a place where 0:06:55.156 --> 0:06:57.556 you could have gone back to the MD PhD program 0:06:57.556 --> 0:07:00.676 for a while. Yeah. I took a leave of a 0:07:00.676 --> 0:07:05.356 continuous leave of absence for probably over five years, probably 0:07:05.396 --> 0:07:07.676 more than I should have, until the point where a 0:07:07.716 --> 0:07:09.956 lot of my friends are like, are you really, are 0:07:09.956 --> 0:07:13.236 you really gonna go back? And finally the medical school 0:07:13.316 --> 0:07:15.276 is like, you're not really going to come back, let's 0:07:15.316 --> 0:07:18.756 just terminate your leave of absence. But it was in 0:07:18.796 --> 0:07:21.036 the first few years a really important safety net for 0:07:21.116 --> 0:07:25.076 me that gave me the psychological safety to really take 0:07:25.156 --> 0:07:28.556 a risk and really pursue a new idea that I 0:07:28.596 --> 0:07:30.796 don't know if I would have otherwise. And I think 0:07:30.836 --> 0:07:34.476 that's so important. I think for universities to provide is 0:07:34.556 --> 0:07:36.956 that to recognize there can be more than one track 0:07:37.756 --> 0:07:40.956 for people to do really excellent science and make an 0:07:40.996 --> 0:07:44.716 impact more than just becoming a professor. And sometimes that 0:07:44.796 --> 0:07:49.156 psychological safety is what's needed to help people find their 0:07:49.236 --> 0:07:54.876 ultimate calling too. By the ways, so far, By the way, 0:07:55.956 --> 0:08:02.716 what's a very brief definition of computational biology. It's really, 0:08:02.716 --> 0:08:04.236 at the end of the day, in my view, just 0:08:04.316 --> 0:08:09.196 the use of computers and data sets to understand and 0:08:09.316 --> 0:08:13.956 biology better. By the way, what happened to that molecule 0:08:14.836 --> 0:08:17.396 that you were testing in mice in grad school? That 0:08:17.476 --> 0:08:22.316 seemed useful? I don't know. It's a good question. Actually, 0:08:26.396 --> 0:08:29.436 I think the project was taken on by someone else, 0:08:30.196 --> 0:08:33.956 but I'm not actually completely sure. So, Okay, you leave 0:08:33.956 --> 0:08:37.276 grad school, you start a company you in fact now 0:08:37.636 --> 0:08:41.436 have taken You have a bunch of molecules that you're 0:08:41.436 --> 0:08:45.076 working on, and that seemed promising. But there's one that 0:08:45.236 --> 0:08:49.196 is in clinical trials now right to treat als Luke 0:08:49.196 --> 0:08:52.156 Gary's disease, a terrible disease that is very poorly treated. 0:08:53.556 --> 0:08:57.676 And I thought that we could talk about the story 0:08:57.716 --> 0:09:00.356 of that molecule of that drug as a way to 0:09:00.436 --> 0:09:03.276 understand the way your company works. Can you just sort 0:09:03.316 --> 0:09:08.396 of take me through the life of that drug? So far? Yeah, absolutely. 0:09:08.516 --> 0:09:13.196 I'll start off just by talking about als and why 0:09:13.356 --> 0:09:16.796 it's been so hard to discover the right therapy, and 0:09:16.836 --> 0:09:19.836 then you know why how we did that differently. So, 0:09:19.956 --> 0:09:22.956 as you might know, LS Luke Garrig's disease is a 0:09:22.996 --> 0:09:27.316 really horrible disease. What happens is that these neurons called 0:09:27.356 --> 0:09:31.876 motor neurons start dying, and most patients experience paralysis and 0:09:31.916 --> 0:09:35.556 then death, usually within three to five years of diagnosis. 0:09:35.596 --> 0:09:39.676 A very fast progressing disease, and there really aren't any 0:09:40.036 --> 0:09:44.716 meaningfully effective treatments that really slow or stop the disease today. 0:09:44.876 --> 0:09:48.316 So a very horrible disease with a horrible prognosis and 0:09:48.356 --> 0:09:52.036 no available treatments, and why it's been so hard I 0:09:52.076 --> 0:09:55.676 think to discover really effective treatments is really just the 0:09:55.716 --> 0:09:59.596 complexity of the disease, and really any disease of the brain, 0:09:59.636 --> 0:10:02.076 the brain is the most complex organ in the body. 0:10:02.556 --> 0:10:05.116 So you end up having a lot of drugs brought 0:10:05.156 --> 0:10:07.596 into clinical trials that worked in mice. I always like 0:10:07.636 --> 0:10:09.636 to say we've cured LS or can There are many 0:10:09.636 --> 0:10:12.996 diseases in mice a thousand times, but none of them 0:10:13.036 --> 0:10:16.316 have really worked in humans. So what we did differently 0:10:16.476 --> 0:10:19.676 was we started from day one by collecting data from 0:10:19.676 --> 0:10:24.316 over a thousand ALS patients as well as controls, and specifically, 0:10:24.396 --> 0:10:28.156 we collected samples of brain tissue as well as spinal 0:10:28.236 --> 0:10:32.196 cords from these patients that actually passed away from ALS. 0:10:32.196 --> 0:10:36.796 So you got samples from a thousand patients who had 0:10:37.156 --> 0:10:40.796 died of ALS. How did you do that? So what 0:10:40.836 --> 0:10:43.356 we've done over the last seven years is we've signed 0:10:43.356 --> 0:10:49.396 partnerships with over twenty one different brain banks, hospitals, labs, 0:10:49.516 --> 0:10:53.956 academic centers worldwide that collect these brain tissues. They're usually 0:10:53.956 --> 0:10:56.996 donated from patients that have passed away from the disease 0:10:57.076 --> 0:11:00.676 and whose families want to contribute to research. Could So 0:11:00.716 --> 0:11:05.596 step one basically is get tissue samples from real patients. 0:11:06.036 --> 0:11:08.596 And you said controls as well, right, So tissue samples 0:11:08.596 --> 0:11:10.556 from healthy people as well, so that you can use 0:11:10.596 --> 0:11:14.076 them as a basis of comparison. You have the samples, Now, 0:11:14.196 --> 0:11:18.236 what's step two? So step two is that we put 0:11:18.236 --> 0:11:22.156 an enormous amount of effort into quality controlling these, So 0:11:22.236 --> 0:11:26.716 that's a big underappreciated step. They can be very noisy samples. 0:11:27.316 --> 0:11:31.196 And then step three is that we sequence them, so 0:11:31.236 --> 0:11:35.916 we profile, what is the expression of all twenty thousand 0:11:35.956 --> 0:11:40.196 genes in the genome, and we also sometimes do DNA sequencing, 0:11:40.276 --> 0:11:44.236 we look at genetic mutations. We also have a clinical 0:11:44.276 --> 0:11:47.636 information about that patient, how long did they have the disease, 0:11:47.756 --> 0:11:52.636 when did they die? And that makes for a very rich, 0:11:52.716 --> 0:11:56.316 multidimensional data set, and that gives us essentially a global 0:11:56.396 --> 0:12:01.076 snapshot of what happened in that patient. H okay, and 0:12:01.156 --> 0:12:05.516 you and presumably the sequencing that you're doing on the 0:12:05.556 --> 0:12:08.996 patient's tissue samples, you're doing the same sequencing on the controls, 0:12:09.236 --> 0:12:12.476 the samples from healthy people. So now you have this 0:12:12.756 --> 0:12:17.636 very large data set. What's the next step. So then 0:12:17.716 --> 0:12:20.556 you have this snapshot of what happened, and the tricky 0:12:20.596 --> 0:12:22.916 part is to figure out what caused it. I often 0:12:22.996 --> 0:12:26.436 liken it to a plane has crashed, right, You're looking 0:12:26.436 --> 0:12:28.676 through the rubble and you want to figure out how 0:12:28.716 --> 0:12:31.316 the plane crashed and how that information can be used 0:12:31.356 --> 0:12:35.956 to prevent further planes from crashing. So that's when our 0:12:36.076 --> 0:12:39.396 software engineers and data scientists as well as machine learning 0:12:39.396 --> 0:12:43.196 scientists come in and we have algorithms essentially to integrate 0:12:43.276 --> 0:12:46.596 multiple data types all the way from the RNA, so 0:12:46.716 --> 0:12:50.436 how the genes were expressed to genetic mutations to essentially 0:12:50.476 --> 0:12:54.836 create a map of disease biology, and within the map 0:12:54.876 --> 0:12:58.636 our networks of genes that are all interconnected that we 0:12:58.716 --> 0:13:02.036 believe cause disease. And so I like to think about 0:13:02.036 --> 0:13:04.836 it like when you're looking through a plane crash the rubble, 0:13:04.876 --> 0:13:07.556 you want to find the black box, which I'll help 0:13:07.596 --> 0:13:10.676 you figure out the cause of the disease. And by 0:13:10.716 --> 0:13:14.236 having all the information, we essentially locate the black boxes 0:13:14.276 --> 0:13:17.156 of disease, the targets that are really at the center 0:13:17.196 --> 0:13:20.556 of those networks, and then we design drugs against those 0:13:20.596 --> 0:13:24.196 targets that we believe can reverse disease. It seems like 0:13:24.956 --> 0:13:29.036 differentiating between correlation and causality in this particular setting would 0:13:29.036 --> 0:13:31.116 be really hard, right, Like to use the plane metaphor, 0:13:31.116 --> 0:13:32.596 if you had a bunch of planes that crash and 0:13:32.636 --> 0:13:34.836 a bunch that hadn't crashed, you might say, oh, like 0:13:35.276 --> 0:13:38.116 the wings were off all the ones that crashed, and 0:13:38.156 --> 0:13:40.476 that's why they crashed. But actually the wings came off 0:13:40.516 --> 0:13:42.596 because they crashed, right, and it was something else that 0:13:42.596 --> 0:13:45.156 caused the crash. I feel like that would be I mean, 0:13:45.596 --> 0:13:49.756 an obvious problem. Yeah, that might be hard. To solve Absolutely, 0:13:49.796 --> 0:13:51.356 you hit the nail on the head, and actually the 0:13:51.356 --> 0:13:53.756 plane metaphor is a really great one here. For one 0:13:53.796 --> 0:13:57.036 of the biggest challenges with looking at tissue from a 0:13:57.036 --> 0:13:59.956 patient that already died is that you're getting the crash right. 0:13:59.956 --> 0:14:03.676 You're not seeing video of before the crash. You're really 0:14:03.716 --> 0:14:06.396 getting the crash. And the challenge is how do you 0:14:06.436 --> 0:14:09.836 figure what caused the crash versus as well was just 0:14:10.156 --> 0:14:14.436 the effect of the crash, like a burned wing, etc. 0:14:15.116 --> 0:14:18.196 And one of the ways we do that is we 0:14:18.396 --> 0:14:21.436 combine different data types. So we found that looking at 0:14:21.476 --> 0:14:24.996 one type of data, for example, just RNA data is 0:14:25.036 --> 0:14:28.516 in particularly helpful, but it's actually looking at where do 0:14:28.556 --> 0:14:31.596 you get convergence signal that pulls through multiple types of 0:14:31.676 --> 0:14:35.836 data to start revealing more compelling signal. So as an example, 0:14:35.956 --> 0:14:38.916 we look at genetic data as well. So genetic data 0:14:39.036 --> 0:14:42.476 is useful for looking at cause versus effect because it 0:14:42.476 --> 0:14:46.036 contains information about genetic mutations that you were born with 0:14:46.116 --> 0:14:50.196 as a baby that then lead to increased risk later 0:14:50.276 --> 0:14:52.676 in life for a disease. And that's kind of nature's 0:14:53.116 --> 0:14:56.636 human experiment for really cause and effect. And when we 0:14:56.796 --> 0:15:00.556 layer that on that information on with the RNA data. 0:15:00.756 --> 0:15:03.636 It actually gives us information about how the genetic drivers 0:15:03.676 --> 0:15:07.556 are acting in these functional pathways, which is a big 0:15:07.596 --> 0:15:10.236 issue actually with just looking at genetic data on its own. 0:15:10.476 --> 0:15:11.836 So I wish I had a better I wish I 0:15:11.836 --> 0:15:13.436 had a way to actually string that through to the 0:15:13.476 --> 0:15:18.876 plane metaphor. But and there's a time for leaving metaphors behind. 0:15:20.676 --> 0:15:24.676 Your company uses AI in drug discovery. I appreciate in 0:15:24.676 --> 0:15:27.556 a certain way that you haven't said AI yet, but 0:15:27.716 --> 0:15:29.876 also I don't want to not talk about it. I 0:15:29.876 --> 0:15:32.316 mean in the sort of figuring out what's going on 0:15:32.356 --> 0:15:35.996 in this step? Is that well, is that the first 0:15:36.036 --> 0:15:38.396 instance in this process where you're using AI? Is it? 0:15:38.396 --> 0:15:41.556 We're talking about that here? Yeah, I mean, I think 0:15:41.556 --> 0:15:45.076 AI is a really broad term for any kind of 0:15:45.196 --> 0:15:48.836 process where the computer is learning from something. So there 0:15:48.876 --> 0:15:53.036 are all sorts of applications of AI in this entire process, 0:15:53.076 --> 0:15:56.836 for example, how we're integrating the data sets together, how 0:15:56.836 --> 0:16:03.316 we're inferring what are the central nodes or the key targets. 0:16:04.036 --> 0:16:07.276 I would say the most classical use of A on 0:16:07.356 --> 0:16:09.476 the way that most people think of it is then 0:16:09.516 --> 0:16:11.836 once we have this network of say one hundred genes, 0:16:12.276 --> 0:16:15.156 how do we actually find what the cause is? How 0:16:15.196 --> 0:16:18.356 do we find what is the hub or the right 0:16:18.396 --> 0:16:21.996 target to hit to turn off or on all hundred 0:16:22.036 --> 0:16:24.556 of those genes. And that's where machine learning and AI 0:16:24.756 --> 0:16:31.876 comes in handy. In a minute, Alice explains how this 0:16:31.956 --> 0:16:35.596 actually works in the case of the ALS drug verges 0:16:35.676 --> 0:16:46.436 working on. Now now back to the show. So okay, 0:16:46.476 --> 0:16:49.076 Alice and her colleagues at Verge have collected all these 0:16:49.116 --> 0:16:52.876 tissue samples from ALS patients. They've used the samples to 0:16:52.916 --> 0:16:56.716 generate this huge data set that shows genetic variation and 0:16:56.876 --> 0:17:00.036 changes in how genes are expressed, along with lots of 0:17:00.076 --> 0:17:03.876 clinical data about the patients, and then they build these 0:17:03.876 --> 0:17:07.956 basically these AI models to try to figure out where 0:17:08.156 --> 0:17:12.276 in this complicated biological process that's happening in this disease, 0:17:12.636 --> 0:17:16.276 where they should try to intervene with a drug, Basically 0:17:16.316 --> 0:17:19.396 where they should try and target a drug. I think 0:17:19.396 --> 0:17:22.356 of this oftentimes, like if you think of a map 0:17:22.356 --> 0:17:24.596 of all the airports in the US, you want to 0:17:24.596 --> 0:17:29.156 figure out how to go after the hubs like Chicago 0:17:29.316 --> 0:17:32.316 or New York. You don't want to go an airport 0:17:32.356 --> 0:17:34.556 in Kansas or I will wouldn't be very effective at 0:17:34.596 --> 0:17:38.796 stopping airplane travel in the country. So there's a lot 0:17:38.836 --> 0:17:42.476 of different pieces of information that we collect to then 0:17:42.716 --> 0:17:45.156 infer what are the best genes that are not only 0:17:45.276 --> 0:17:49.196 central within this network, but also there's independent evidence of 0:17:49.436 --> 0:17:54.276 a disease causal effect or a relationship to disease. And 0:17:54.356 --> 0:17:59.836 so you do all that in this instance, and what 0:17:59.916 --> 0:18:03.076 do you figure out? So what the algorithms spit out 0:18:03.076 --> 0:18:06.236 is essentially a ranked list of targets, all right, So 0:18:06.276 --> 0:18:08.596 these are ranked list of targets that are predicted if 0:18:08.596 --> 0:18:12.996 we could dry them, would restore that network back to 0:18:13.116 --> 0:18:18.156 levels of healthy people and potentially slow or stop the disease. 0:18:19.156 --> 0:18:20.916 And then what we do is we take those targets 0:18:20.916 --> 0:18:23.396 and we start testing them in the lab, all right, 0:18:23.436 --> 0:18:25.356 So we actually what is kind of cool about the 0:18:25.396 --> 0:18:27.556 platform is we get all these targets from human brain 0:18:27.596 --> 0:18:30.716 tissue and we also can test them in human brain 0:18:30.796 --> 0:18:34.156 cells in the lab. So you get a list it's 0:18:34.196 --> 0:18:37.436 basically genes to target. You either it says upregulate or 0:18:37.676 --> 0:18:40.156 make this gene express more or make this gene express less. 0:18:40.236 --> 0:18:44.236 Is that basically what the AI is out putting exactly, Like, 0:18:44.836 --> 0:18:47.916 so how long in the instance of this ALS drug. 0:18:47.916 --> 0:18:50.636 How long was the list? More or less, our initial 0:18:50.676 --> 0:18:55.716 set of targets was twenty two high confidence targets, and 0:18:55.756 --> 0:18:59.916 then we actually then generated another chut choosing updated data 0:19:00.076 --> 0:19:03.916 of about thirty more targets as well. And what was 0:19:04.036 --> 0:19:07.396 really striking when we tested these targets is that when 0:19:07.436 --> 0:19:10.596 we tested them in the lab, we found that on 0:19:10.836 --> 0:19:14.716 average over sixty percent of them, though more recently actually 0:19:14.756 --> 0:19:17.756 around eighty percent of them actually validated in the lab, 0:19:17.836 --> 0:19:22.036 so they actually protected ALS patient cells from dying, which 0:19:22.076 --> 0:19:25.356 is very high. So we're really excited that we're actually 0:19:25.356 --> 0:19:29.556 seeing very robust validation of the computational predictions, at least 0:19:29.596 --> 0:19:33.596 in the lab. Okay, so you have this list, you're 0:19:33.636 --> 0:19:37.436 testing it, something like half of them seem promising, you said, 0:19:37.436 --> 0:19:41.996 sixty percent seem promising. What happens next? Okay, So what 0:19:42.036 --> 0:19:44.796 happens next is that we so we test them in 0:19:44.836 --> 0:19:48.796 these human brain cells. We understand the mechanism. One of 0:19:48.836 --> 0:19:52.116 the really interesting findings from this ALS program and specific 0:19:52.356 --> 0:19:54.236 is that when we looked at the network that we 0:19:54.316 --> 0:19:57.596 found in these patient spinal cords, we found a new 0:19:57.636 --> 0:20:02.276 cause of disease that was previously unknown so most of 0:20:02.316 --> 0:20:04.836 the hypotheses in ALS, where many of them to date, 0:20:04.876 --> 0:20:08.556 have really been focused around these protein aggregates, these clumps 0:20:08.556 --> 0:20:11.316 of proteins that we can easily observe by ie that 0:20:11.396 --> 0:20:13.636 you see in ALS patients. Right, A lot of them 0:20:13.636 --> 0:20:17.716 are observational hypotheses. But what we found by looking at 0:20:17.716 --> 0:20:21.036 a deeper cut of the data is actually, at baseline, 0:20:21.716 --> 0:20:25.716 most of these patients actually had a baseline dysfunction in 0:20:25.756 --> 0:20:28.556 their life csomal pathway, which I like to call the 0:20:28.596 --> 0:20:32.436 garbage disposal pathway. It's what is critical to clear out 0:20:32.556 --> 0:20:36.596 junk from the cell. And because patients were at baseline 0:20:36.796 --> 0:20:40.156 vulnerable to these toxic insults, it wasn't so much the 0:20:40.196 --> 0:20:42.876 protein clumps that were directly causing it. It was because 0:20:42.916 --> 0:20:47.076 they're already vulnerable to these clumps of proteins that their 0:20:47.076 --> 0:20:51.476 cells started dying. And is the idea that the gene 0:20:51.476 --> 0:20:57.196 you're targeting is causing the cell's garbage disposal to not work, right, 0:20:57.236 --> 0:20:59.476 Like you're trying to fix the garbage disposal by targeting 0:20:59.476 --> 0:21:04.516 this particular gene. Yeah, it's a central regulator of that pathway. 0:21:04.796 --> 0:21:06.756 And it was also a target that was ranked I 0:21:06.836 --> 0:21:08.516 think it was ranked number one or number two on 0:21:08.556 --> 0:21:12.756 the list. So just to be clear, how how do 0:21:12.836 --> 0:21:16.916 you get from you know, so you have fifty or 0:21:16.956 --> 0:21:21.476 so things to test, fifty or so targets, something like 0:21:23.476 --> 0:21:26.716 thirty of them seem promising. How do you decide which 0:21:26.716 --> 0:21:30.956 of those thirty to proceed with? Yeah, so that's a 0:21:30.996 --> 0:21:33.236 great question. We get asked that a lot. I think 0:21:33.316 --> 0:21:36.556 at that point it's a strategic decision. Right, you were 0:21:36.596 --> 0:21:39.796 a startup, Right, we have to be able to develop 0:21:39.836 --> 0:21:43.436 things quickly and capital efficiently. So we were lucky in 0:21:43.476 --> 0:21:47.036 that sense that one of the top targets was also 0:21:47.036 --> 0:21:51.076 a target that already had where the path to developing 0:21:51.116 --> 0:21:54.796 a drug was relatively smooth, A lot was known about 0:21:54.796 --> 0:21:57.876 that target. We could start doing chemistry and designing molecules 0:21:57.916 --> 0:22:01.956 relatively easily, and the target itself had actually been tested 0:22:02.356 --> 0:22:07.476 in the clinic for other diseases, not als, but things 0:22:07.516 --> 0:22:10.716 like Crohn's disease and surrounds, so we did know there 0:22:10.796 --> 0:22:15.596 was some safety data around hitting that target. We do 0:22:15.676 --> 0:22:20.036 then for targets where we can't develop all of the targets, right, 0:22:20.076 --> 0:22:23.196 we can only take focused bets for targets where there's 0:22:23.236 --> 0:22:25.916 a bit more technical risk, Right, It might be a 0:22:25.916 --> 0:22:28.996 bit more exotic. People don't really understand how it works. 0:22:29.876 --> 0:22:33.156 There's not a lot of tools out there to really 0:22:33.196 --> 0:22:37.276 develop drugs against it. That's where we might partner with 0:22:37.316 --> 0:22:41.916 a pharma company to develop those targets. And we have 0:22:42.236 --> 0:22:45.156 such a collaboration with Eli Lily where we developed our 0:22:45.196 --> 0:22:48.676 als target, but actually Lily has the opportunity to essentially 0:22:48.756 --> 0:22:53.636 take you targets number three through twenty two plus and 0:22:53.796 --> 0:22:57.076 choose four of them to develop themselves. Oh interesting. So 0:22:57.156 --> 0:23:00.316 in that way, you're essentially laying off the risk to 0:23:00.396 --> 0:23:03.396 this giant pharma company that can afford to make more bets. 0:23:04.236 --> 0:23:07.756 I'd say we're distributing the risk and we're allowing us 0:23:07.756 --> 0:23:12.236 to really capitalize on the entire opportunity all of the targets, 0:23:12.276 --> 0:23:15.236 because it's impossible for any small startup to do, you know, 0:23:15.436 --> 0:23:18.676 thirty different programs. And it's actually in line with what 0:23:18.796 --> 0:23:20.596 a lot of pharma companies are looking for. A lot 0:23:20.636 --> 0:23:23.916 of pharma companies are looking for. What is that novel 0:23:24.116 --> 0:23:26.396 target that no one else is working on that's kind 0:23:26.436 --> 0:23:29.636 of unexpected, Where if we could really get a competitive 0:23:29.716 --> 0:23:32.356 edge in here, this would be really meaningful for a 0:23:32.436 --> 0:23:37.836 position within within drug development in the next ten years. Well, 0:23:37.876 --> 0:23:42.636 and I mean it also seems compelling because even though 0:23:43.156 --> 0:23:46.116 this seems like a more promising way to do drug development, 0:23:46.596 --> 0:23:51.596 drug development is hard enough that anyone candidate drug is 0:23:51.636 --> 0:23:55.396 probably not going to work, right. Yeah, An any biotechniqus 0:23:55.396 --> 0:23:57.316 to be able to have a pipeline and the ability 0:23:57.316 --> 0:24:00.956 to withstand I think some failures because I think it's 0:24:01.036 --> 0:24:03.796 unrealistic to expect one hundred percent of what you try 0:24:03.836 --> 0:24:07.876 will work. But that doesn't reflect on the technology itself, 0:24:08.556 --> 0:24:11.516 and that can be something unfortunate in biotech, where you know, 0:24:11.516 --> 0:24:16.116 if the first thing fails, everyone's all can be. It 0:24:16.156 --> 0:24:18.876 can be tempted to say, oh, the technology didn't work, 0:24:18.916 --> 0:24:21.676 but in reality, you think about how many different drugs 0:24:21.716 --> 0:24:24.516