Episode 140: The Growing Role of AI in Healthcare, Biotech, and Drug Development

In today's episode of BIO from the Bayou, we're going to be talking about AI, the good, the bad, and the slightly nuanced. You're listening to BIO from the Bayou, empowering biotech leaders with the tools, assets, and expertise to improve healthcare outcomes.

Each episode provides insights and information to expand your network, advance your product line, and help your company thrive.

Serving the entire Gulf South, but recorded in the heart of New Orleans. Welcome back to BIO from the Bayou. I'm your host, Dr. Elaine Hamm, the executive in residence at Tulane School of Medicine. And today's guest is Dr. Rafael Rosengarten. He is the founder and CEO of Genialis, a precision oncology company utilizing AI to help the more promising molecules become life-saving medicines. Welcome to the show.

Thank you, Elaine. It's a pleasure to be on.

It's good to see you. It's been a while since I have seen you.

I think I ran into you at JP Morgan and, and—

Oh yeah, on the stage.

Insisted on carrying your suitcase across the stage.

Yeah, you, you were very, very helpful. I love someone that when they come to say hi to me, that they also Helped me carry my luggage too. So I have much appreciated that. So, well, am I gonna see you at BIO as well? Are you, are you headed down there?

Absolutely. Yeah.

All right.

We've got ASCO coming up this weekend, so probably this episode will air after that, but ASCO first, then BIO, and then I get a break for the summer.

Oh, nice. Where's ASCO this year?

Chicago. Always Chicago. It's the one that sticks. It's always Chicago. That's right. Yeah.

Yeah. Oh, well, it's so great to see you on, on the podcast and I mean, we've, known each other through our various organizations, including the Alliance for Artificial Intelligence in Healthcare, which you were one of the founders of. So I obviously, we're gonna talk a little bit about AI, but knowing you and I will probably cover a gamut of different topics.

Absolutely.

Well, let's talk a little bit about the progress of AI in healthcare. We've been involved in the space for a while, you longer than I have, but it's developed quite a bit. So let's talk first about your company and your origin story as it relates I'll try to make this long story not too long for the air.

My introduction to machine learning, which is a branch of artificial intelligence, goes all the way back to 2011. So there are certainly people in the world who've been doing it longer, but in healthcare, that's pretty old school. I'll go ahead and say that's old school. I moved to Houston in 2011 from the Bay Area where I had been doing a postdoc. As any good story starts, I followed a girl. To a new city and started a new postdoc at Baylor College of Medicine. And it was one of these just kind of pure academic happenstance things. My biological background, my lab science background, because I am definitely a bench scientist, has always been in sort of weird, just-so story systems, not especially translational, but where the biology is really neat and where there's a lot of natural history and just a way to kind of be a kind of classic and intellectually curious. So I was working in a lab on this organism called Dictyostelium. And again, academic happenstance, they had this longstanding collaboration, like decades old, with a data science lab in Europe. Dr. Blaž Župan at the University of Ljubljana, Slovenia, was in the faculty of computer science. And for decades they would do data science over in Slovenia to analyze the data from Houston, from Baylor College of Medicine, and build sort of prototypes of software that the lab scientists could use, et cetera. My first year in the lab, Blaž came over from Slovenia for a whole year for a sabbatical and brought some graduate students with him. And his graduate students included a woman named Marinka Žitnik. Marinka is, in my estimation, and probably a lot of other people's, has become one of the top 2 or 3 academic AI for healthcare scientists. She's at Harvard now. She postdoc'd in Yuri Leskovich's lab at Stanford. Yuri Leskovich, it turns out, was one of those other top scientists in this field. They're both Slovenian. They both did their graduate work in the same lab where my co-founder did his graduate work. So you kind of see where this is going. I got this academic introduction, got introduced to this new mathematics where they could build models, mathematical models that would predict outcomes in the lab. We'd go and test those outcomes and we're like, holy shit, that just saved me 2 years of lab work. Right. And this was kind of like right around the time that CRISPR was becoming a thing, but we didn't actually have CRISPR harnessed yet in our system. So I fell in love with this. I drank the Kool-Aid with both hands. I got a, Fellowship to fund my research that insisted that I actually learn about AI, at least learn enough to have conversations. So I did this interdisciplinary fellowship through the Gulf Coast Consortium, which was hosted at Rice, but with 6 different organizations. Dr. Lydia Kavraki from Rice was sort of overseeing that, and it was a real springboard for me. I was not going to be an academic scientist. I kind of knew that.

Yeah.

I knew it for a lot of reasons. I was too much of a journeyman, studied too many different systems without ever latching onto one. And maybe more importantly, I was head over heels for this person who's now my wife, and she had a real career. And I was like, all right, whatever, wherever your career takes us, I'll figure it out. And so entrepreneurship seemed like a better way of being mobile, but also being able to remain a dilettante and harness this Kool-Aid that I'd been drinking and say, how can we bring this to bear on biomedical research? So one of these graduate students again became my co-founder. He had also come over to Baylor to kind of do an industry postdoc to figure out what purpose this technology should serve. And so in 2015, I think it was NIH, might've been NSF SBIR grant, but I wrote an SBIR grant to build one of these AI models of certain cancers in which we would take all of the world's knowledge of that cancer, all of the data we could find, and we would map it into a single AI system that could be trained to understand to learn from all this information, and then could be pointed at any prediction task you might have had, whether you wanted to discover new drugs or figure out patient populations, et cetera. Now, I was unaware at the time, we might not have even had the word yet, but what I'm describing to you is now what we think of as a foundation model. And foundation model is like, it is one of the buzzwords in the AI space, right? LLMs are foundation models, right? ChatGPT is a foundation model.

Yeah.

But you also see these becoming hyper-specialized in healthcare. You have foundation models of X, Y, and Z. So without realizing it, based on largely on Marinka Zitnik's work, which I was mostly a spectator of, but got to participate in a little bit, I knew that this was the thing we should do. Teach computers the corpus of knowledge and then make it do its work for us. We didn't get funded for that grant. Hmm. So we got a good score, but we didn't get funded. And so the company kind of, we started the company and we took some other twists and turns, basically chasing commercial traction. and built some great software that allowed us to start aggregating data. We knew that data aggregation was going to be the bottleneck. In 2015, there was not big data in the life sciences, right? We didn't have large corpuses of omics data like Tempus and CARIS. We didn't have these large compendia. It was frankly a small data era.

Yeah.

But we've harnessed this technology to aggregate enough small data to start doing real AI. Jump ahead now more than a decade. Genialis has built the first AI foundation model of cancer biology trained on largely the transcriptome and all the other sort of knowledge that feeds into that. And we do exactly that. We work with pharma to try to figure out who are the right patients to get your drug and what are the right drugs to give to these patients. So we try to really streamline what is still the probably most pernicious bottleneck in drug development, which is clinical trial failure.

Sure. Well, you started on cancer. What cancer Specifically, were you looking at at the beginning?

The grant was, if I'm not mistaken, was for medulloblastoma and chordoma, which are, are brain cancers and, and, well, and CNS cancers. And that was pretty nice. The reason why, frankly, was—

Yeah, why?

We had become, as a company, we'd become friends with a researcher named Charles Lin. He was a junior faculty at Baylor at the time. In fact, his first day on faculty was my last day as a postdoc. And my PI was like, before you walk out the door, stop by this office and introduce yourself. Charles had just come from Dana-Farber where he had trained under under Jay Bradner, who's now SVP of Research at Amgen. Charles came in with a mandate to reinvigorate computational biology in the Department of Molecular and Human Genetics at Baylor. He had come out of a lab at MIT where he helped sort of design and partly engineer a software system for this. And so he, he basically partnered with us to build this software for computational biology. So Charles happened to not specialize in, but be an expert in Medulloblastoma and chordoma. And so we were going to kind of team up, which was the reason why we focused there. It was for no good purpose, except there was a lot of unmet need and an opportunity. What we've done since, what Genialysis has done though, we don't have a cancer type specialty. We've built a truly pan-cancer system. And the way we did that is by not so much focusing on what do we know about cancer A or B or C, but rather Can we teach the AI to understand ground truth biology? Can we teach it to understand all of the different facets in kind of a reductionist sense? Can you break cancer into its component biological parts, right? Or could you represent all of cancer with a big box of Legos, right? So we built all these Legos.

Yeah.

And each one of these Legos represents some aspect of cancer biology. Maybe it's pathway signaling, like MAP kinase pathway activation, or maybe it's Oncogene dependency, like KRAS dependency or c-Myc dependency. Maybe it has to do with synthetic lethality or, or some multi-gene effect like DNA damage response. Maybe it's to do with the tumor microenvironment, the milieu of the tumor, right? So we start, we have Lego bricks representing the immune system and the vascularization system around the tumor. We have a substantial enough Lego box now that it doesn't matter what cancer, and it also doesn't frankly matter what target your drug hits. We can build a predictor that does the job by reassembling it from a subset of Lego bricks.

Well, you mentioned you started out with small datasets, and I think that's really interesting because when we think about AI and when I think about it in particular, I think of very large datasets, and that's how we've trained a lot of our models, quote unquote. But how did— what's the difference when you start with something that's a small dataset?

This is a great question, and I think my attitude about this is maybe a little bit apocryphal in the field, right? So first of all, everyone wants to talk about big data because we spend a lot of money on data. And so we want the data to be big because it was expensive. The truth is, I would argue that almost all data in life sciences is small data. Even when you look at large compendia, right? Even if you consider, again, let's just take Tempus as a good example because it's a well-known example. They have somewhere between 400,000 and 500,000 transcriptomes. And I'm focusing on transcriptome because that's my company's favorite data. Genialis likes RNA-seq. Caris has even more. Cool. So here's your instinct then. You've got a half million transcriptomes. So your instinct is to throw that at a computer and say, find me the patterns, right? Machine learning AI is really good at finding these hidden patterns.

Right.

To my knowledge, we've yet to build a biological model where you've just thrown a lot of data at it, where it's learned general principles enough to be applied widely. You can build specific models that way. But if you want to build a specific model, that's the wrong dataset for it because the data are general. That's all of cancer and tons of different clinical history. It's messy. It's real world. It's— the mess is a feature, not a bug. So what we've done at Genialys is really embrace this idea that biological datasets are relatively small. And here, you know, think of a typical clinical trial, right? Phase 1A and B, you might get to 50 patients. Phase 2, maybe 150. Phase 3, 400 or 500 patients across lots of trials. Maybe you've got several thousand. But, you know, if you're measuring the transcriptome, that's like 20,000 stable genes. So the data matrix is the wrong shape. You've got 20,000 possible features for your model, those are your genes, and 2,000 patients, right? It should be the other way around. You want more observations than features, because otherwise the model isn't going to generalize. It'll overfit. So we've embraced this small data problem.

Yeah.

And there's an acronym called HDLSS, high dimensionality, low sample size. I didn't make it up. Doesn't roll off the tongue. It's just a term of art. But this problem of high-dimensional, low sample size is part of why these great technologies like next-gen sequencing, omics, all the different kinds of -omes and multi-omics, why that's hard to use in modeling. Because you would need a massive dataset to be able to just throw omics at a computer and get it to come up with a good generalizable answer.

Right.

So instead, Genialysis spent a decade building tools for integrating small datasets and for interrogating small datasets.

Yeah, right.

And one of the amazing emergent properties of our foundation model, of our system, is we can build and train AI predictors with a sufficiently few number of those Lego bricks I've described to you.

Yeah.

Each one of which is a machine learning model, right? So each one of which already embraces the sort of power of the nonlinear pattern recognition that ML gives you, but gives you a simple score on a biologically defined And you could take like 5 of those and train a clinical predictor on 50 patients.

Yeah.

And get a really good predictor that you could take to the bank. You can take it to a trial, you can take it to the FDA. This predictor is gonna work and you can get away with— put it another way, your clinical trial data, if you're a drug company, is gold. But yeah, it's partly gold because it's so precious, it's so rare. There's not a lot of it. You don't want to run into a 50-patient dataset trying to train a predictor with 20,000 genes. But if you've got 5 vectors, you can do it. And so we've kind of, we haven't solved, but we have a solution to this high-dimensional, low-sample-size problem. And this allows us to help pharma build really important, really informative therapeutic intelligence models using their precious early clinical trial data, but to build these before you've dosed your first patient and to lock them into clinical utility after your first phase of clinical development.

So why RNA then?

There are good, you know, sort of good reasons and opportunistic ones. The good reason is we think the best compromise of genotypic and phenotypic information, right? So you've got the genetic information there. And in fact, we have patent-pending technology to call DNA variants, so mutations out of the coding regions of the RNA. So you don't even need to do DNA sequencing if you're running our software. You get all, you get all the coding region variants, which is most of the ones we know what to do about. But you also get all this phenotypic information. What is the dynamic state of the tumor, right?

Yeah.

So rather than relying just on DNA, which is sort of the state of the art for sequencing-based diagnostics, is, you know, you do a DNA panel or a whole genome, that tells you what's possible, but the RNA tells you what's happening. Now, you could argue that protein or metabolites, you know, is getting you closer to that phenotype still.

Yeah. That's true.

But with the sacrifice of some robustness and resolution, RNA-seq is now a 20-year-old technology, more than 20-year-old technology, maybe. Gene expression certainly is a 25-year-old technology. It's clinically robust. It's becoming cheap. It's reimbursable, at least in the United States. I think it's going to become reimbursable everywhere as the price comes down.

Yeah.

And so we think that, yeah, there are much shinier analytes out there. There are things that are fancier and flashier and more cutting edge. But if you want something that, again, gives you the mix of information richness with the commoditization you need to take it into clinical devices. We really like RNA. We've got R&D programs with lots of other analytes, but our sort of commercial stage stuff is kind of focused on what we think we can get through FDA and EMA today.

Speaking of all of those many years you mentioned, you've been in AI and drug development for quite some time. What has Changed since you first started?

There's been a convergence, a combination of the data getting bigger. So the data are approaching big data, and certainly in the chemistry space, that's true. There have been people working in life sciences doing algorithm development and really pushing the envelope. You know, hat tip to folks like Brandon Allgood and team at Numerit, which was sort of one of the OG first generative chemistry companies. They were too early to market. Like there really wasn't much of a market for what they were doing. No one believed it. Companies like Insilico Medicine and our friend Alex Zhurankov, who got patents on things like reinforcement learning and stuff that's sort of used in other fields, came out of some of his work. But also all the big frontier labs, the technology groups, Googles, and so forth. They just are really good at AI. So the algorithms have gotten a lot better. And we can think about this maybe in terms of pre- and post-ChatGPT moment.

Mm-hmm.

And here I'll use the Alliance for AI in Healthcare, the global nonprofit that you and I've been involved in, as an example. When we started in 2019, we had our first sort of our inauguration, our first board meeting at JP Morgan that winter. That was kind of our stage. And for years, for several years, or at least a couple years leading into COVID, we tried to get platforms and there was really nowhere to speak, or we had to kind of by hook and crook, we were on the side stage. And sort of COVID happened and then ChatGPT happened. And as soon as AI was on everyone's desktop, on everyone's phone, in everyone's lexicon, as soon as it was in everyone's living room, when your mom's talking about it, my mom's talking about it, like then all of a sudden pharma like, oh shit, this is real. And so now, right, like Nvidia is the hardest talk to get into at JP Morgan.

Like, right, right.

Yeah. So, I guess the point is saying the industry as a whole, like with the rest of the world realized that, okay, sure, there's AI hype undoubtedly, but there is so much substance. And this was the premise that we founded the Alliance on is this is real. Sure, there's hype. We're going to help you separate the wheat from the chaff, but we need to till the fields for when this lands because there are going to be drugs. that are AI-generated. There are going to be diagnostics that run on AI. There are going to be healthcare decisions. And this is all true today. So 7.5, 8 years ago, we were like, we need to start doing some work. We need to educate the regulators, educate the policymakers, think about standards, you know, at least build some thought leadership around what does the world look like when everything changes. And like any good overnight success, it was decades in the making, right? But all of a sudden— Absolutely.

Absolutely.

All of a sudden, It was the buzzword, but buzzy because it had arrived, not because it was pure speculation.

Well, so it went from, I mean, it seemed it had its sexy moment, but I feel like there's an interesting, as the public, as you said, like even your mom's talking about it, even my mom's talking about it. And now we sort of have pushback in a lot of ways. There's distrust of it. There's concerns about data centers. There's concerns about all of these things that Are the other side of that coin of AI. It's one thing, I think, when we talk about AI in derisive terms, if we think about generating the human image of your dog, for example, like, give me— yeah.

Versus like— What is the value?

Yeah. What we're trying to do in drug development and in healthcare to make sure that patients are getting the drugs they need, the help they need, the access that they need. And it's an interesting, I don't wanna call it a battlefield, but it's interesting these days because it feels like it's even changed within the past 2 years, year and a half?

It's a really good question. I'm the CEO of a tech company. I consider myself a Luddite. At the end of the workday, I put my phone on a shelf, close my computer, and I go out to the garden and I dig in the dirt because I need something tactile and real and analog, right? I am not one of those sort of like optimist techno enthusiasts in the way you might expect for someone who works in a business. And if you think again about the Alliance for AI in Healthcare, we weren't founded just to promote AI. founded to advocate for the, quote, responsible adoption of AI, right? Big underscores under that. You know, my attitude is in healthcare broadly, healthcare's a huge umbrella.

Yeah.

You know, responsible means, at least for now, for sure, human in the loop and maybe even more than human in the loop, right? Like, let's continue to treat AI as a tool. And this is with the recognition that especially as we move into an agentic era where you can build guardrails, like all of the agentic stuff we run internally in my company for all of our processes, We've taught it the guardrails, right? We've taught it to raise red flags if certain kinds of data—

Sure.

—are entering. Like, you can make the AI your buddy. You can make it help. You can turn it into a quality control machine. But I think the architecture in healthcare needs to be very thoughtful around these guardrails and these quality controls. I think there needs to be, and this has been my soapbox for years, there needs to be a tacit and explicit recognition that there are gaps in the data and biases in the data. So if you're building healthcare AI, you better damn well appreciate that maybe in your dataset there's an underrepresentation of certain— there could be an underrepresentation of women, for crying out loud, right?

Yeah. It's all built on white men.

Right. So the first step is to acknowledge that that may exist. The second step is to build a tool to quantify how bad it is. And then the third step is to figure out how to fix it. So in healthcare especially, let's think, this is where we can play and where we can make a difference about that. More broadly in society, I don't know that I'm in a position to push back against the, like, dramatic waste of resources for all of these silly uses of this technology that burn a ton of energy, burn a ton of resources, et cetera. And that's one major issue, right? The resource consumption is something I, you know, I don't have a solution for it, but I try not to use, for example, ChatGPT for things that I don't have to use it for. Although even I am falling— That was a good question. Like, in my work life, like, I've started trying, like, things that take used to take me half an hour. If I need it, if I want it done in a minute or 2, or I just don't want to do it at all. Like there are certain things that, holy hell, that's great. And I'm starting to justify that to myself. I no longer write my podcast briefs by hand, right? Like I, and they're really, they're really thorough.

Ian's laughing and smiling. Yeah. 'Cause I'm pretty sure that's also something that he might do.

A little bit, a little bit once in a while.

That's okay. No judgment. No judgment. Judge-free zone. Yeah.

No, but, but again, it's one of those things where I can get a more thorough podcast brief. It'll take me 10 minutes to read and be a better interviewer than if I were to spend 30 minutes on it and like kind of burn out and just like do the minimum. So things like that, is that a good use of the bottle of water and the 2 trees it took to make it? I don't know. My bigger concern is different than resource use, and maybe I should be focused on resources. My bigger concern is I'm the parent of 2 children, right? They're almost 12 and almost 7, summer birthdays. They're elementary school age.

Yeah.

The older one, the middle schooler, all of his schoolwork is on a laptop. He carries a laptop around every class, carries it home. Most of the assignments the teachers are giving him are written by AI. Most of them are graded by AI. He takes that as a sign the teachers don't care, and so he cares less. If kids are not taught to write, I don't know how they're going to be taught to think, right? Like, writing is a form of thinking. It's a crucible through which we distill our thoughts, right?

Yeah.

So these are my concerns. And maybe I'm just like this old, old man yelling at the clouds and shaking my fist at the clouds at this point. I don't know. But I am genuinely worried about people becoming dumber.

I agree.

And I'm worried about raising kids in a world in which it's so uncertain, like, what are gonna be the kind of useful skills going forward? And, and what I actually start thinking about is that the useful skills are gonna become more the softer skills. Empathy, listening, critical thinking. Not that critical thinking is a soft skill, but like—

Soft does not mean not difficult.

Right. So, for example, my wife and I have embraced our son's interests. He's joined the debate team at school. He actually is the debate team. He was the only one from the whole school initially signed up. And, but it's something like this, like, all right, cool. Here's an activity where Even if you were to use AI to prep for your debate notes, you've got to— You still have the moment. You flex that muscle, right?

You have to.

You've gotta think about communication. You gotta think about framing arguments. So this is where my concern with AI comes in. I think we're gonna get it right in healthcare because it's already a regulated environment. It's already— yeah, absolutely. I think that if we lean into the existing guardrails, really kind of bake that in, worry about data biases, data gaps, it would be nice to have certain, like, slightly different political winds to kind of take it more seriously. But anyway, society—

There's a whole other podcast.

Society and like general direct-to-consumer markets are smoking with you got 'em kind of thing. We're going to burn the trees and the bottles of water, and that's going to take a different kind of effort. And I think we are seeing it as political grassroots opposition to the data centers and things like that is kind of where that's going to come from. There may be economic forces too, like maybe this is a house of cards built on sand, or to mix a metaphor, right? Like the whole thing may crumble under the pressure of like If we invest 8 gazillion dollars in data centers and we have to recoup that in revenue, is that even possible? But the piece I'm worried about is this sort of what happens when kids don't learn how to think.

That's my biggest fear, honestly, is neurocognition. I have found, I have caught myself becoming overly reliant on it, and I have really thought about that on a long-term basis. I'm in neurospace. I have a family history of dementia. And so like, I want to make sure that I'm still pushing myself and also touching grass, as you said. I mean, I think that's an important piece of it that that we need to be aware of, and maybe less human in the loop and more humanity in the loop. Is that a new thing I can say?

Please.

I like that. I'll go with that. But yeah, I think about that all the time. I mean, there's things— do I want to interrogate an Excel spreadsheet, take out all of these things, and have ChatGPT do that? Yes. But I worry about the creative piece. And a lot of science, most of science, there's a, there's a huge chunk of creativity. And that's what I worry about, is are we going to lose our creativity as we think about science? So, yeah.

That's a brilliant point too, to create. I didn't even mention the creative angle. I was thinking about like learning to, to read and think and write, but creativity's huge and, and it does impact things like science. You know, this, it does. My scientific heroes are mostly also artists and— They are same musicians. And you know, there's a reason why these skill sets tend to kind of coalesce in people, right? Because it's left brain, right brain kinds of interactions. And so I, I am worried about, I don't know which side of the brain's gonna be hurt worse by AI, but I'm worried that it may sever the connection between the two. Right. I think it's the severing that I worry Yeah.

Um, I literally am picturing the neurons right there. So, well, this has been fascinating. I think we could go on for another hour and we probably didn't even touch like half of the questions we had prepared, but that's, we knew that was gonna happen. But I really appreciate chatting with you and I will see you at BIO this year. I won't have any luggage for you to have to carry, I promise.

You can, that's cool. For those of you who don't know, Elaine has carried my water for years with the Alliance for AI in Healthcare. So, so it's the least I can do is to wheel a suitcase across the street.

I'm still, I have stopped doing the 990 tax forms literally this morning. So yeah, still, still ongoing. Still ongoing.

Well, I don't know if this is going to make it in off the editing room floor, but at BIO, the plan is for a reception. I think it's the Tuesday evening at BIO, co-hosted by Connect, which is the new parent organization for the Alliance for AI in Healthcare. We folded in January. And Bullpen.

And Tulane.

And Tulane.

New Orleans Biotech Innovation Center and my company, Cadenza. It is going to be a hell of a party. I've—

Yeah.

I'm planning on shipping some Mardi Gras beads. That's something that, Ian, I'll have you help me follow up on.

Of course.

And it's going to be a Fat Tuesday theme. Of course, we didn't have a chance to do our usual BIO On the Bayou, From the Bayou, one of those receptions last year. It just was chaotic. So if you're going to be at BIO, please join us for our Fat Tuesday reception with Connect AIH, Bullpen, Tulane, Nobic, and Cadenza.

Cadenza, yeah.

That's a lot of beads to be thrown around, which will make you earn actually at this reception. But great to see you again. I can't wait to see you in person.

Yeah. Well, don't correct me with a good time. I'll see you, see you in about a month.

Yeah, sounds good. All right. Thank you.

Thanks, Celine. Talk soon.

Thank you so much for joining us, and we'll be sure to keep an eye out on the work that Jamie Ellis is doing and your work in the future. And of course, we always have our information about our guests of the company in our show notes. Thanks for joining us for Bio from the Bayou.

Thanks for joining us for Bio from the Bayou, and we hope you'll join us again.

To never miss the latest updates in biotech, hit the subscribe button now.

We'll catch you on our next episode of Bio from the Bayou.

This podcast was edited by Resonate Recordings.