The Legacy of Stanford’s Biomedical Informatics Program
Harry traveled to the San Francisco Bay Area this summer, and while there he interviewed the co-founders of three local data-driven diagnostics and drug discovery startups, all of whom participated in the same graduate program: the Biomedical Informatics Program at Stanford’s School of Medicine. Joining Harry were Aria Pharmaceuticals co-founder and CEO Andrew Radin, BigHat Biosciences co-founder and chief scientific officer Peyton Greenside, and Inflammatix co-founder and CEO Tim Sweeney. The conversation covered how each company’s work to advance healthcare and therapeutics rests on data and computation, and how the ideas, skills, connections each entrepreneur picked up at Stanford have played into their startups and their careers.
Radin’s company, formerly known as twoXar, models pathogenesis computationally to identify potential drug molecules, shaving years off the drug development process. Radin developed Aria’s core technology, a collection of proprietary algorithms for discovering novel small molecule therapies. The algorithms incorporate system biology data, disease-specific data, chemistry libraries, and more than 60 separate AI methods to sift through molecules with known chemistry to find those with novel mechanisms of action and favorable safety profiles absorption properties. Whereas traditional drug discovery methods have a 1-2% success rate after 4 years, Aria claims its approach has a 30% success rate after just 6 months. It has a pipeline of at 18 drug candidates in areas including kidney, lung, and liver diseases, lupus, cancers of the liver and lung, glioblastoma, and glaucoma. Radin holds MS and BS degrees in computer science from Rochester Institute of Technology, studied computational biology and medicine through the Stanford Center for Professional Development, and was formerly an advisor to several venture capital firms and startup accelerators.
Greenside started BigHat to combine wet-lab science and machine learning with the goal of speeding up the design of antibody therapies. BigHat’s lab consists of numerous “workcells,” each of which cycles through multiple tests of a given set of antibodies synthesized from in silico designs. Assays characterize each antibody variant for traits such as yield, stability, solubility, specificity, affinity, and function. Machine learning algorithms determine how mutations affected each of these properties and feed this learning back into a new set of designs for the next round. The company says this approach allows it to identify therapeutic-grade antibodies faster than traditional bulk screening techniques (in days rather than weeks or months). Greenside is a computational biologist with a PhD from Stanford, an MPhil from Cambridge University, and a BA from Harvard. Silicon Valley Business Journal named her to its 2021 list of “Women of Influence in Silicon Valley.”
Sweeney co-founded Inflammatix to develop a new class of diagnostic tests that—rather than searching for a specific bug—“read” the host response of a patient’s immune system for clues about the cause and severity of an infection. The problem, as Sweeney originally saw it, is that traditional tests can only detect infections once a pathogen has spread to the bloodstream, meaning that doctors often guess incorrectly about whether a patient needs an antibiotic, or which one they need. Inflammatix is built around the idea that the human immune system has evolved targeted responses to different kinds of infections and other diseases. These responses are complex and vary according to age and setting, but by analyzing mRNA samples from multiple, diverse cohorts, the company believes it can identify a “reproducible signal in the ‘noise’ of multiple datasets.” Inflammatix is developing a cartridge-based system called Myrna for use in emergency rooms, urgent care clinics, and outpatient clinics that can screen for acute bacterial infections, viral infections, and sepsis in 30 minutes. Sweeney is a physician and data scientist who earned an MD/PhD from Duke and then trained as a general surgery resident at Stanford.
Please rate and review MoneyBall Medicine on Apple Podcasts! Here’s how to do that from an iPhone, iPad, or iPod touch:
1.Open the Podcasts app on your iPhone, iPad, or Mac.
2.Navigate to the page of the MoneyBall Medicine podcast. You can find it by searching for it or selecting it from your library. Just note that you’ll have to go to the series page which shows all the episodes, not just the page for a single episode.
3.Scroll down to find the subhead titled “Ratings & Reviews.”
4.Under one of the highlighted reviews, select “Write a Review.”
5.Next, select a star rating at the top — you have the option of choosing between one and five stars.
6.Using the text box at the top, write a title for your review. Then, in the lower text box, write your review. Your review can be up to 300 words long.
7.Once you’ve finished, select “Send” or “Save” in the top-right corner.
8.If you’ve never left a podcast review before, enter a nickname. Your nickname will be displayed next to any reviews you leave from here on out.
9.After selecting a nickname, tap OK. Your review may not be immediately visible.
Harry Glorikian: I’m Harry Glorikian, and this is MoneyBall Medicine, the interview podcast where we meet researchers, entrepreneurs, and physicians who are using the power of data to improve patient health and make healthcare delivery more efficient. You can think of each episode as a new chapter in the never-ending audio version of my 2017 book, “MoneyBall Medicine: Thriving in the New Data-Driven Healthcare Market.” If you like the show, please do us a favor and leave a rating and review at Apple Podcasts.
Harry Glorikian:Home base for MoneyBall Medicine is the Boston area. It’s one of the world capitals for biomedical innovation and the digital revolution in healthcare. So I don’t have to venture far to find great guests.
But obviously Boston isn’t the only capital for biosciences innovation. This summer, during the brief break between surges in the coronavirus pandemic, I escaped to the San Francisco Bay area. And while I was there, I got a lesson about the considerable impact created by one particular Bay Area institution. Namely, the Stanford School of Medicine’s Biomedical Informatics program, or BMI for short.
BMI trains students how to use and adapt computational methods like machine learning to solve hard problems in biology and medicine. And a remarkable number of BMI alumni have fanned out into the world of life science startups. On today’s show you’ll hear from three of them. We’ll talk about the work their companies are doing now and how the skills and connections they picked up at Stanford have played into their careers.
The first guest, and the person who helped to organize the group interview, has actually been on the show twice before. His name is Andrew Radin, and he joined me in November of 2018 and again in August of 2020 to talk about his Palo Alto-based company Aria Pharmaceuticals, formerly known as twoXar.
Aria uses a collection of proprietary AI algorithms to discover new small-molecule drugs for a range of diseases. In traditional drug discovery, years can go by between the identification of a new drug candidate and testing the drug in animals. Radin says Aria’s AI can reduce that time to just weeks.
Andrew kindly recruited two of his fellow Stanford BMI alumni for our conversation. One is Peyton Greenside, the co-founder and chief scientific officer at BigHat Biosciences in San Carlos, California. The company combines wet-lab science and machine learning to make it easier and faster to design new antibody therapies. And again, the leap forward is that BigHat’s rapid cycle of antibody design, synthesis, and characterization vastly speed things up, reducing the time required to identify new therapeutic antibodies from months to just days.
And our final guest is Tim Sweeney. He trained as a surgery resident at Stanford and then founded a company to tackle one of the biggest problems in acute care, namely how to diagnose infections faster and more accurately. The company is called Inflammatix, and it’s building a device that emergency departments and outpatient clinics can use to rapidly analyze messenger RNA in patients’ blood to screen for sepsis and other kinds of infections.
All three of these companies are benefiting in different ways from the computational methods their founders studied at Stanford. And they’ve got some great stories to share about how their time at BMI convinced them that future progress in medicine and drug discovery would depend on data above all else.
We originally planned to meet up in person for this interview. But we switched to Zoom at the last minute out of concerns over the Delta variant. So without further ado, here’s my talk with Andrew Radin, Peyton Greenside, and Tim Sweeney.
Harry Glorikian: Well, hello everybody. And welcome to today’s show.
Tim Sweeney: Thank you.
Peyton Greenside: It’s great to be here.
Harry Glorikian: Yeah, it’s, it’s great to have all of you here. For everybody listening and watching, we were actually supposed to do this in person, but unfortunately the Delta variant sort of threw a monkey wrench in that whole process. So I reserve the right that we can do this in the future and actually get together when this whole thing is over, like normal human beings.
Each of you are working on super exciting things. Different companies, focusing in different areas. And I know you all know each other, so I’m going to step back one second and say, if you had to give a brief description of your company or pretend you don’t know each other, where we’re at a cocktail party and you’re going to give me two or three sentences about what you’re doing and why it’s interesting, how would you sort of do that? And Andrew, since you’re the ringleader that sort of helped bring this group together, I’ll throw it out to you first to sort of get going.
Andrew Radin: Well, that’s a lot of pressure, but certainly like, our description I think is pretty simple. We are a preclinical stage pharmaceutical company. And we happen to have a proprietary artificial intelligence platform that’s discovered all the assets that we have under development. And these days we have 18 programs, 18 different disease areas where we’ve got new experimental medications and we are working on progressing those new inventions to the clinic and ultimately to FDA approval.
Harry Glorikian: Peyton?
Peyton Greenside: Hi everyone. I’m Peyton and one of the co-founders of Big Hat Biosciences, and our mission is to improve human health by making it easier to design advanced antibody therapeutics. So we actually do that through a combination of a high-speed wet lab and machine learning techniques in order to very iteratively design and improve antibodies until they meet unmet patient need. And it’s been a lot of fun. Then we’ve been founded since 2019.
Harry Glorikian: And finally, Tim.
Tim Sweeney: thanks for the opportunity, Harry. Inflammatix was founded about five years ago, spun out of Stanford along with, of course, Aria and Big Hat. We are designing novel diagnostics focused on acute care and critical illness needs. So we basically have a data analytics platform that allows us to decode certain signals of gene expression within the immune system. And then for those of you watching, I’ll show you, we have a cartridge that allows us to sort of implement that in a 30 minute point of care diagnostic setting.
So our particular focus is basically bringing precision medicine into acute care settings, the hospital, the clinic, the ICU, where sort of historically there hasn’t been a lot of diagnostic innovation.
Harry Glorikian: Interesting. That’s funny because I actually, I wrote a a textbook on how to commercialize novel diagnostics a few years ago. Because you know, unless you’ve been through the ringer, you may not know all the different pieces.
But you guys now all know each other right? Now, that may not surprising because we’re in Silicon Valley, and I’m actually in Berkeley right now, but that’s close enough. And drug discovery companies and tech companies are all swimming around each other. But your connection is a little bit deeper. I mean, you guys all went to Stanford together. So this is not necessarily a commercial for Stanford, but it’s, that’s pretty interesting that three CEOs of data-driven, you know, healthcare companies out of the same class, whoosh, come out of Stanford. So how did you, how did you guys meet at first?
Andrew Radin: Well, and I would say we’re not the only ones to—it’s just, you know, the people that happen to be in front of you today. It was funny. So, right before this, I sent a panicked email, because I didn’t want to say something that wasn’t true. I was like, Peyton, you were in this class, weren’t you?
Peyton Greenside: Yeah. I don’t know if I was Andrew’s TA or if we’d all actually been in the same class. But I think our Stanford journeys all started, it sounds like, the same year. Same time. And we all were taking translational bioinformatics, which was a course taught by, I believe, Atul Butte who I think, you know, really brought to fame the idea of big data for biology, what you can draw out a very large data sets and drawing insights. So we were all in the same class and with many other people, as Andrew said, and it was a lot of fun. And I think it was the start of long journeys for all of us than in a similar vein.
Andrew Radin: And it was a place for…I think what was awesome about that class, again, not to be an advertisement for the coursework, but it was kind of my characterization of the class was, you basically learned how other people use data science to solve some medical mystery, like across the spectrum. And so the, the purpose of learning all that was to just kind of fill you full of ideas of things that you could do. And then the kind of the capstone of the class was a final project where you basically had to come up with something, right? And so you were just sort of primed with all this like super interesting sort of research on how other people had approached very different problems in the space. And for me, it was just the source of lots of interesting ideas that then, you know, helped me ultimately create what’s the technology behind our company today.
Tim Sweeney: It is remarkable how much came out of Stanford biomedical informatics. Though, I mean, to Andrew’s point, there are, there are a number of other CEOs that came through in that sort of in maybe a five or seven year stretch, all out of the same program. And I think a lot of it had to do with that, yes, this one particular class had all the different applications of data science sort of across the spectrum of life sciences, but they also attracted people like that. Right? I mean, everyone on this call has a very different background before Stanford BMI. And I think that was part of what made that culture so special is that it ended up being a real team sport, whether your background was medicine or business or math or computer science or bio-engineering or anything else, learning a technique from A, and applying it into area B, I think, was a pretty successful way to grow innovation.
Harry Glorikian: I feel like as a venture guy, I should be standing at the exit door and just sort of saying, you know, “What’s your idea, what’s your idea,” screening as they’re coming out the door.
Peyton Greenside: Well, you know, some folks have also become venture capitalists.
Harry Glorikian: That’s true.
Peyton Greenside: Yep.
Harry Glorikian: So was there anything in particular that you guys, interests or questions or discussions that you sort of bonded over that sort of brought you together? I mean, even, even as just friends that decided to keep in touch?
Andrew Radin: Well, I think it’s probably different for different people. I think the first real interaction I had with Tim, you know,the details escape me, because this is almost 10 years ago now, but I remember, he’s a medical doctor, right? He’s got a MD and a PhD if I’m, if I’m not mistaken. And so my, you know, I’m a hardcore computer scientist. That’s my background. And so back in those days, I was rapidly learning all I could about medicine and biology. And I don’t remember the topic, but I do recall him helping me after class was something that wasn’t just quite, you know, sitting in my head correctly. And I remember thinking like, what a nice dude, to, like, you know, kind of take some time and give me like, you know, a little private tutoring.
And then and then if I recall afterwards, you said, yeah, so I’m trying to do this stuff with some clinical data. Can you help me with this sort of stuff? Which if I remember correctly, I never actually helped you. I was talking about, oh, I might be able to help you. And then eventually you said, “I figured it out. I don’t need you”
Tim Sweeney: I said I needed to build a web scraper. And I said, I have no idea how to.
Andrew Radin: Oh yeah, I have totally done that. Lots of times. So yeah, something like that. That’s how the conversation started with Tim, which was sort of to the point about having very different backgrounds, You know, with Peyton, I don’t really recall the first interaction. I remember we were in a journal club, maybe with Russ and you were talking about some stuff, but I think the more I got connected to her was around the time she was working on her defense and I actually went to her PhD defense. And I have this BS detector that sometimes go off a little early, right? When people make a statement, I’m like, “I don’t know about that.” We’re sitting in her defense and every time she said something that made me, do one of these, like, “Wait a minute,” she instantly resolved that in the next sentence. I was like, “Okay. All right. That’s cool.”
Peyton Greenside: Okay, that feels good. Fortunately, fortunately.
Andrew Radin: You don’t have to pass my scrutiny obviously, but yeah, I think that led to a number of kind of interesting conversations as she was contemplating, you know, what to do next. She was moving through her career, but yeah, I think that the interactions are very, very different for each person. At least that’s my view, but I don’t know if you guys have different memories.
Peyton Greenside: Yeah, I think what’s, what’s interesting, I mean, just generally I agree with that. And I think one of the most interesting parts of BMI, as Tim said, is just the backgrounds that everyone has. And I also come from the kind of applied math, computer science background, and there’s this kind of fascination of what you can do with computational skills in biology. I think to me, a lot of the conversations were around where do I even apply this to? I think people sort of think of computational biology as a, maybe sort of a niche, small field at the intersection of maybe somewhere where biology meets, I guess, you know, statistics, computer science and math. But it’s so broad and it’s so vast. And I think most of the, I say the most exciting conversations I’ve had are, you know, we work in immunology, you know, you’re a clinician, you work with clinical data. How do you apply these tools? The most daunting but fun task upon showing up at Stanford with such an incredible ecosystem here is, where do you even focus your attention? Where should you work? There’s too many exciting opportunities to pick. And I think some of the fun conversations I remember also having a Tim, with a more clinical background, is what’s actually useful? You know, I want, I want to do something useful and sort of try to figure out, you know, where this, you know, where are you can actually kind of apply your time to the most impactful problem. It was a lot of fun.
Andrew Radin: And I think, Tim, it’d be great for you to share. I mean, when we first met, I’d asked you kind of like, what were you doing there? What your story was? I can’t remember the words back then. But you basically said like, “Look, I’m a surgeon,” if I recall, “I’m trying to save people’s lives and I’m just thinking like, is there a better way? Can I like just, you know, do something that’s going to have a much larger impact? And I don’t know what that is yet.” I know I’m wildly paraphrasing what you said, right. But I’m thinking about like what that could be. And I think. You know, when I met you, you were sort of on the hunt for figuring out where to apply, you know, kind of the, the skill set.
Tim Sweeney: I think that the everyone shows up with their strengths and weaknesses. Mine certainly was the summer before the program actually started, I had to take, you know, basic courses in computer science and linear algebra. And I remember, I mean, I literally went from my last overnight call at Santa Clara Valley Medical Center, running two ICUs, to the next morning CS 106x. Which, because it was the summer, was filled with all the high school students that are just total whiz kids, like 16 year olds, and they’re like, you know, we’re learning like order of operations or something and they’re raising their hands and I’m like desperately trying to write down like, oh, if n means….
You know, and obviously Andrew and Peyton were among the folks that sort of helped me on the basic science side of things. But I think that the story about sort of getting the question right is absolutely correct. And I remember actually the time that I knew I was in the right program was maybe two or three months in to training. They used to have these like sort of work in progress talks, and it was like, you know, Wednesday or Thursday or something, you bring a lunch. And somebody was talking about this thing that sounded very, very cool to me. It was all about how you could, you could program a system to learn new knowledge on its own. And it was like, you know, generalized AI for health data. And I was incredibly impressed. And, and the first example that was given was like, you know, so we’ve sifted through all of the billions of data points. And I have discovered—he stumbles over the drug name—I’ve discovered that plopacapagril, by which he meant clopidogrel, is associated with bleeding events. And everyone goes, “oh.” And I put my hand up, like, “That’s an anti-platelet medication.” And he looks at me and I’m like, “the point of that is that it thins the blood.” He looked at me and was like, “So bleeding is a known side effect?” Totally crestfallen that people knew this already. Like, he had no idea. I was like, I do have something to contribute, so it’s good. It’s a good merge.
Harry Glorikian: Yeah. So, you know, Tim, you’re running a diagnostics company you know, Peyton and Andrew you’re running what I’ll lump together as drug discovery companies in different markets, different regulatory processes. You know, I’m sure there are common challenges to life science startups in the valley. What are some of the biggest challenges that you guys see? Is it scalability? Is it finding the right people? Is it finding the right investors? Where do you guys see your challenges?
Andrew Radin: And I would just say for a little clarification to Peyton’s point about there’s so many different problems. Even though Peyton and I are both in the business of creating new medicines, we couldn’t be any more different. We’re a small molecule company. She’s a large molecule company. If you know what that means. You know, I’m making motorcycles, she’s making trucks. Like, we’re just, we’re just, we’re just doing completely different things. To your question about like, kind of what are the very similar things, we’re not really even competing with one another from that perspective.
But I think, to answer your question, at least from my viewpoint, you kind of have to do all those things. I think, you know, in startups, everything has to work. You can’t sort of have any one thing that doesn’t function and whether that’s the science or the fundraising or the team or all of those things, if you’ve got a problem in any one of those areas, it can be life-threatening to the company.
And so I think part of the experience for the entrepreneur is sort of, you know, because your time is limited and your resources are limited is sort of finding a best fit to try to solve, you know, or, or to maximize all of those problems simultaneously. And I would say all the things that you’ve listed, they all at various points in the company, they’ve been critical and it’s more of a juggling act rather than “Geez, all you need to do is just knock it out of the park, on, you know, financing and who cares about anything else?” We know lots of stories where that hasn’t gone well. Or you knock it on the park on an exceptional team, and the other things don’t come together. So, you know, from my standpoint, all of that stuff has to work.
Peyton Greenside: I think my answer continues. I think one of the things I, and what many people who just find, I would say many scientific, inquiries fascinating, is just what to work on that. And I have the same problem now, you know, I think it happened when I went to Stanford and happened you know, postdoc and have it happened now.
And, in the context of my company, wo we basically have a platform that can work on engineering any protein. We work on antibodies, but really can be anything. So, you know, we have this landscape. There are tons of diseases with unmet need. There’s sort of tons of opportunities for the type of therapeutic protein you would use, whether that’s a standard antibody, monoclonal IgG, sort of a next generation antibody. And so we always have to decide, you know, what, what are the programs gonna be? What are you going to go after? What’s the modality? And I think at the crux of it, like you know, for a drug discovery company, is what is the shape of your company. But our platform is so broad that basically we can work on so many things. And I, once again, by myself faced the same problem, which is okay, like, you know, where should we focus our attention? And that’s been really fun. This is getting maybe more of Tim’s background, but so we’re learning more about the clinical side of things and where that need is and where that pairs with our technology. But I agree with what Andrew said, nothing really can go shortchanged, but that’s been the same theme, I would say just now in a different vein.
Harry Glorikian: Yeah. I mean, I think about this as a balance of dynamics where you’re at different stages at different points, depending on where you are in the development cycle. And you need different people and different issues become a problem at different points or maybe become more acute at different points. But you know, all of you guys have one theme in common, which is why we’re on the show together. It’s data and some form of machine learning or other, you know, part of artificial intelligence that’s being applied to find something valuable or identify some valuable piece of information that can make something actionable. It’s sort of a big question, but how do you employ machine learning and AI in what you’re doing in each of your businesses? Because I think of these things as like I have a toolbox and then I have to apply that tool in a very specific way with a specific set of knowledge that can feed it, where I can get an output that I’m looking for. And so each one of you, like you said, Andrew, you’re, you’re working on the motorcycle, she’s working on the big truck, and he’s trying to make sure that everybody gets diagnosed and not, not ends up in worse than they already are. So how are you each of you thinking or approaching this in your own unique way? If you can summarize. Tim, why don’t you go first?
Tim Sweeney: Our tests work by measuring a discrete number of genes within the body. It’s their expression levels. So for instance, for our flagship test inset, we look at 29 different gene expression levels from, from blood. And then of course we have to somehow integrate 29 different levels into actionable information. And so the backend of that is the data science part, the machine learning. So step one is actually choosing what to measure. And then after you’ve chosen what to measure, then it’s training hardened algorithms that turn 29 different things into a score that says, “This person has a bacterial infection.” And then of course doing that repeatably, doing it in a way that is traceable and verifiable. And then all of the post hoc, you know, how is it affected by different demographics? And how has it, in the actual context of care, and of course in the coming years when actually implemented in a health system, how does it impact patients and providers and does it save costs and improve outcomes?
And maybe just since I didn’t get a chance to answer, I think one of the questions about challenges is a lot of times it changes with the application that you’re taking farther. Right? One of the things that we all have in common, I think is that we’re all platform companies. And to, to Peyton’s point, like you can apply that data science platform to a lot of different areas, but each one of those areas has to be taken through a very long development process to actually help a person and the challenges totally change along that development life cycle.
Harry Glorikian: And just for everybody listening—so you developed this product. What is the, so what, what is the impact?
Tim Sweeney: In our case, we decided that we wanted to go after one first indication that would be a big enough hit to make the business matter. We’ve got lots of things we’d like to do in the long run, but sepsis is an area of outstanding unmet need. And the “so what” is right now, if you go in and you’re feeling sick and you see a doctor and you want to know, Hey doc, like, do I need antibiotics? There is literally no test that can answer that question. It’s a guess. So it’s not to say that antibiotics aren’t administered quickly, but as a physician myself, I can tell you that that is it’s a guess at first, and then you have to wait for tests to come back and those tests themselves are imperfect. And so something like 40% of antibiotics are probably misprescribed.
And if you knew in 30 minutes, Hey, this person has a bacterial infection or no, you could greatly simplify care and really improve outcomes. And that’s the premise. But the challenge of course is that beyond the data science, there’s so much that goes into building the product and proving out the clinical data and get it through FDA and then getting it reimbursed and, and, you know, getting it rolled out more broadly, if you want to get to the point where you’ve actually helped a number of people and built a solid business.
Harry Glorikian: When I, in my last company, before I moved on to venture, I, we had a strategy consulting firm and we did a lot of digging into sepsis. That was a big problem, a nut that people were trying to crack, and, you know, if you could crack it, the opportunity is quite significant.
So Peyton, Andrew, how do you guys think about it? Because I’m, I’m thinking manipulating an antibody and sort of tweaking little parts of it until you find the exact fit. [It requires] supercomputing or massive computing.
Peyton Greenside: It’s funny. I actually think that the context in which we all met, which is you know, when I think big data was becoming really popular in medicine is actually a great context, I think, for where Big Hat ended up, and it’s funny, because it’s going to been kind of a long journey—it always happens when I look back, I’m like, yeah, that makes, that makes sense. Right? Based on where I was. We actually put a lot of our attention into integrating the wet lab with the dry lab. And this is actually, you know, with a goal of making big data into what I might call sort of smart data or agile data, which is that the idea of back in the day when first, I would say you got tons and tons of really large data sets. And you can sort of mine them, or you can look for trends. You can sort of just figure out something, you know, interesting relationship between gene expression and patient outcome. And I kept throughout my career feeling frustrated by being handed the dataset and sort of having to just mine it and not having kind of, you know, ownership of being able to say, “I want to look here, I want more data here.” Right? You’re sort of handed a really large data set and you’re, a passenger in this dataset that has already been generated. You cannot modify it. That’s kind of the fixed dataset. And, you know, as a computational person, that, that you’re often the second person, like a wet lab or experimental lab is making the data, then you kind of get it right. And so, you know, throughout I would say, especially in my time at Stanford this was very much the case, where I was felt kind of trapped in being given a data set that I didn’t actually design, but I could sort of mine.
And so at Big Hat we’re basically trying to now put computation in the driver’s seat and kind of change that paradigm. We’re actually now, instead of just getting one large data set that you design up front, you acknowledge that biology and the science are very iterative, right? As as you said, you sort of start with an antibody sequence, but, you know, would you stop there? If you could just make one tweak, maybe you’d make it, you know, 10x better, 100x better with two. So how do you enable it? How do you want to enable that very rapid cycling? And so we view this as kind of the intersection of how closely can a lab and the computational side interact and how can they inform each other? How can you one learn from the other? And so we actually enabled a computational person to design an antibody on Monday and in a few days you synthesize, purify, characterize the antibody and kind of understand, are you moving in the right direction or are you not? And repeat, and then repeat it and repeat and repeat. So you don’t get kind of stuck in the fixed data set again.
So it’s really attractive for a lot of ways, right? There are a lot of reasons you kind of can end up in a really good regime and it’s big data or sort of area, but, you know, there’s kind of a lot of lost opportunity in terms of being able to kind of be very agile and move toward something that looks promising and then iterate more. And the goal is that that will allow us to enable types of antibodies they don’t even exist today because you can’t engineer them that easily. You’re kind of are stuck with a fixed format. So that’s been really fun. And so we’ve been spending a lot of time designing the wet lab to kind of support the machine learning side and data science side from the ground up and, and vice versa.
And so it’s a pretty unique sort of set up. And I think I like to think of it as sort of smart data, right? You’re thinking really closely about what should I generate that will be helpful and can use that to inform how you redesign the next dataset and improve your antibody every time in our case.
Andrew Radin: Yeah, it’s interesting to hear the different stories. You know, I think all of us are kind of taking the approach that, you know, what data sources and what artificial intelligence allows you to do is to take real world data and then make some prediction under uncertainty. You know, with the expectation that prediction is potentially better than what you could, what you could do with other methods.
And so, you know, kind of tying this back to when I was student and thinking about where are the places I can make a big impact, it was very interesting to me that with very complex diseases there was really no single biomedical measurement that would help kind of unravel the mystery of the biology behind that disease. And therefore could, you know, explain something about pathogenesis that would lead to a new discovery or a new medication as a result. And, you know, part of that coursework in 2.17 was this concept of integrative genomics. This idea of using, you know, different data sources that are all keyed to the same thing, maybe a, a gene or a gene product, and kind of looking for that overlapping evidence.
And there were some great papers that were shown. There was one, I think, by, by Eric Lander in particular, where he was using, GWAS and proteomics and maybe some gene expression microarray data, each of which would give you, you know, like hundreds of quote-unquote “answers” and the real answers in there buried with a bunch of false positives. But ultimately what would happen in this paper is he showed that there was one overlapping gene in all three of these datasets and he ran some assays and determined, indeed that was the key to unlock this mystery.
And that certainly worked well if all of your data sets are sort of keyed to the same thing, but that’s not the reality of biomedical data sets. There’s genomics measures, there’s chemistry measures, there’s phenotypical measures, there’s different patient measures. And unless you’re conveniently measuring them all from the same patient population over time, which is very expensive and very, very time consuming to do, there’s really no easy way to sort of key all these things together.
And my thought was like, “Hmm, maybe, maybe there, there is a way.” And so the technology that I created and ultimately has been expanded upon is taking this concept, the concept that the answer to a very complex disease doesn’t necessarily live in any one measurement or anyone biomedical data set. And if you have the ability to ultimately pull in lots of very diverse—and by diverse I mean statistically independent—data sets across a wide range of biomedical measures and integrate them as a single processing unit, you can ultimately uncover things that other people essentially haven’t noticed before. And then use that, in our case, you know, to do lots of things, but in our case specifically to develop new therapeutics.
So in all of our disease areas, ultimately what this means is we are working on new mechanisms of action. These are, these are new, if you will, new concepts or new understanding of biology in these disease areas and therefore what it means or what the impact is—to your earlier statement—is, we’re going after biology that potentially has a disease modifying effect that others have not approached before. And therefore the promise of the opportunity is to make a significant dent in these very complex diseases.
And so that’s a kind of a high level view of what we do, but ultimately it’s all about, you know, integration of these very different datasets. And then using that to ultimately come up with new experimental medicine that we would explore and experiment with and see what it can mean for patient impact.
Harry Glorikian: Yeah. I think that’s one of the most exciting parts of when I talk to everybody. Assuming the system is designed well, and the data going in is actually good, it’s like, “Wow, I didn’t notice. I didn’t know that that happened. I didn’t know that pathway was involved or this little tweak could make this difference.” And so that’s what I see when I talk to different people that are working in this area. “I just didn’t know,” or “None of the papers talked about this,” or “That’s not what I learned in school.” And so that’s the most fascinating part of these systems where you can identify things faster, hopefully and more accurately, hopefully than you might normally do with a human being. No knock to human beings, all of them are valuable, but it seems the systems move at a different pace and can handle a much broader level of data being input into them.
And so that brings me to the question that Andrew, you and I have talked about. If you had to put a timeframe around it or something is, is this shortening the time to discovery? And I think you and I, the last time we talked, you said to about three years where I can shave off on the front. And then at some point when I have to get to a mouse, I have to follow the normal trajectory of that mouse. But if that’s changed and you you’ve, you’re finding other areas, I’d love to hear it. But Peyton and Tim, where do you see the aha the speed or the financial impact of what you’re doing? You’re doing it because it’s moving at faster or you’re able to identify something that you haven’t, but it’s better than X or Y that’s already being done in the marketplace.
Peyton Greenside: For us actually, this is, I mean, we do do things faster. We do improve on a lot of metrics. But it’s actually, at least for my companyl about designing antibodies that couldn’t otherwise exist. So for example, the standard monoclonal IgG, there are many tools out there to sort of discover initial molecules and optimize them, but you start getting into these kinds of next-generation or kind of Frankenstein antibodies, antibodies that are a tenth of the size, or SCRBs which are these fragments that are part of car T therapies or other formats.
They become more complex and people have trouble engineering them, and you can kind of run your imagination and say, well, if I had the ability to engineer things, what other formats would I conceive? Would I consider, tiny antibodies like cell-penetrating peptides that can get into cells and sort of have all sorts of characteristics? But they’re difficult to engineer.
And so we actually, instead of sort of doing the same thing faster we actually think more about how can we expand the universe of what could be a potential therapeutic protein and how would that solve current patient needs in ways that existing therapeutics do not. And we do that by doing things faster, sort of, and cheaper and, sort of. More smartly. But hopefully that’s what we really care about.
Tim Sweeney: I’d answer probably somewhat like Peyton’s. But if you look at a diagnostics and biomarkers in particular, a lot of diagnostics are about, “Hey, you know, we found that if you measure this one protein that’s useful for health.” So it’s just a very slow process and it’s not optimized. You tend to study things that are obvious because they’re easy to measure. Or like in our field, there’s one protein called procalcitonin that’s sort of the current closest biomarker for whether or not somebody has a bacterial infection, but PCT, as procalcitonin is abbreviated, was discovered 30 years ago and it was originally basically by accident that someone even measured it in someone with bacterial infections, and then it worked pretty well. And you know what I mean, it’s a sort of based on serendipity and it can’t be improved upon it has. However good procalcitonin was yesterday, that’s how good it’s going to be tomorrow and how it’s going to be the day afterwards.
I think the benefit of data science and in diagnostics was really began with cancer, when you had sort of the wonderfully successful tests like Oncotype showing how you could measure signals across complex diseases by integrating things from multiple biomarkers. And a lot of those were designed and there, again, the problem was that they took a long time to develop. And of course they take a long time to actually run, right? I mean, most of them, if you’ve ever had one of those tests done, it’s like a week to send out, you know, you send some tissue to a company, it gets processed. You get your answer seven days later. So one of the things we’re doing differently, one, it has to do with the way that we gather and integrate data sets to empower faster discovery.
And that’s kind of like Andrew. The other is basically the ability to build new answers that haven’t yet existed, sort of more like Peyton. And ultimately the hope is to create a feedback loop where you know, better and better versions of the tests can be slowly released. And so over time, it’s not just that you’re sort of stuck with, “Hey, you know, procalcitonin is as good as it is [going to get].” It’s like, you know, you’re on Insept version five in 2030, and it’s now X percent more accurate. And I think that’s a real benefit to patients.
Harry Glorikian: I want to pause the conversation for a minute to make a quick request.
If you’re a fan of MoneyBall Medicine, you know that we’ve published dozens of interviews with leading scientists and entrepreneurs exploring the boundaries of data-driven healthcare and research. And you can listen to all of those episodes for free at Apple Podcasts, or at my website glorikian.com, or wherever you get your podcasts.
There’s one small thing you can do in return, and that’s to leave a rating and a review of the show on Apple Podcasts. It’s one of the best ways to help other listeners find and follow the show.
If you’ve never posted a review or a rating, it’s easy. All you have to do is open the Apple Podcasts app on your smartphone, search for MoneyBall Medicine, and scroll down to the Ratings & Reviews section. Tap the stars to rate the show, and then tap the link that says Write a Review to leave your comments. It’ll only take a minute, but it’ll help us out immensely. Thank you!
And now back to the show.
Harry Glorikian: So you guys have been doing this for a while. Do you see the promise of big data and AI playing out the way that you thought and or is, or is it different than you thought now that now that you like jumped into the pool and you’ve been swimming in it for a while? Is it fulfilling the dream you had, is it more exciting than you thought?
Andrew Radin: It’s a funny question. Coming from very different industries, you know, looking at where I was 10 years ago, I think I was very naïve about what it actually takes to bring a drug to market. And I think in the very early days of the company, you know, my prior startups, you know, one of them I was in and out in a year and it exited. And there’s no such thing in this industry, to do anything like that. And so, you know, part of it was biased by my prior experience, but I think part of it as well is, sometimes I think it’s also hard to see how far things have moved along. And I think even in Tim’s description is he was sort of talking about, well, you know, this, this was state-of-the-art science, you know, in decades past you know, the work he’s doing today was impossible back then. So, you know, there’s sort of these steady, incremental improvements.
And I, and I think part of what really is happening in the industry is that the things to solve essentially are becoming exponentially harder. For example, for high throughput screening, which is maybe the old way of doing things, to find a hit is exponentially harder. For diagnostic tests or blood tests to sort of detect these nuances, you sort of have to bring in these technologies and these capabilities that are exponentially better at solving those things.
And so I think what happens is, you can therefore characterize it in a different way, you know, is the time faster compared to the old way? Well, of course, because those old ways just don’t have a chance of being able to do these things. Like, is it cheaper? Well, yeah, because those old ways, again, just don’t have a chance. But I think part of it is what is the pace of innovation? And that’s, I think kind of where the rubber meets the road and what is actually possible and what it’s capable of.
And so today, you know, we’re, we talk about having, you know, 18 concurrent disease programs and we’ve got a very small team and we haven’t raised very much money. You know, that would just be flat out impossible 10 years ago. And we still like raise some eyebrows around that, but now, it’s okay. We recognize software is doing a lot of what used to happen in the wet labs. So this, you know, sort of fits within the expectation of what a modern technology company would do in this space.
So I think there’s that other angle of where expectations are kind of catching up with what’s actually been produced. And therefore, you know, at, at some point we become the old technology. Thirty years from now, some next generation we’ll be talking about, oh, those, those slow, painful people that, you know, tried this in the past kind of stuff. And so it’s, you know, each, I think each iteration of innovation has its moment in the sun, if you will. And this is definitely the time for the work that we’re collectively doing.
Peyton Greenside: I think the promise is ahead of us. We’re in an amazing time where I think things are starting to gain traction. We’re starting to get tools and infrastructure, but if I were to say my conception of what machine learning and data science and generally computational power is going to do in biology and medicine, I think it’s just starting.
So I’m excited to see things like AlphaFold. I’m excited to see a lot of these kind of tools and capabilities to be unlocked. But I think, you’re solving a complex problem, right? That protein that you’re affecting is in a cell, it’s part of the tissue, and it’s part of a human, and there’s so many more layers, I think, to consider.
Yeah, we’re making great progress. And I still certainly believe in the potential. That’s why I’m here. But I do like to say, I think we’re at the very, very early days. And as Andrew said, I think it’s going to be fun to see what happens in 30 years. So I’m still very excited, but I wouldn’t say we’re at the accomplishments that I would consider as sort of really demonstrating the cornerstones of machine learning in, in biology and medicine.
Tim Sweeney: I have to agree with Peyton, I think the best is ahead of us. So one of the courses we had to take at Stanford BMI, and I don’t know if you two remember this, was Marc Musen taught this course on ontologies, but part of it had to do with sort of like the history of applications of sort of clinical data systems. And the oldest one, I forget the details, but it was in like, the ’70s. And it was around sort of you know, clinical decision support for therapeutic prescribing. Obviously that system isn’t around today and failed for its own reasons and he sort of walked through all of the failures of systems since then.
And maybe one of the most remarkable things is how, how little AI and machine learning is actually employed in most clinical practice. You know, for all the buzz around computer vision, the AI that radiologists use most is probably their dictation. I mean, it isn’t yet commonplace to have machine assisted radiography reads. And so will that be coming? Absolutely. But the interesting challenges in each successive generation of like, oh, you know, we got pretty close, but it turned out that X wasn’t good enough, or it wasn’t built in the right way to be integrated with workflow or is coming soon, but still needs some regulatory work or whatever else. There’s plenty left to do.
Peyton Greenside: I, I think that’s probably one thing we all experience actually transitioning from academia to industry is, what’s exciting in academia is not necessarily what’s going to be reliable when you really want to make a good drug. So what you might think about it, you’d be like, “Oh man, that’s a really cool model. I’d love to try that, you know, that’s great.” And you kind of go right into industry and you’re like, okay, well this is going to matter. This is, this is going to go to patients. It has to work multiple times. I think it is a very different standard. Right. And so I actually think it’s the right thing. Just because you find something to be very, very cool and kind of, you know, I would say cutting edge, you really want it to work and want it to work over and over again. I think there’s an unappreciated gap between when something is first proposed or conceived of or demonstrated and when you can really make it work at scale, over and over again in areas that matter.
So I think we’re basically in that transition, for, I would say, a lot of these techniques in biology and medicine. Now let’s get to work and practice. Let’s get to work and practice reliably. And now we can start sort of really seeing where we’re going with the needle on really impactful problems. But it’s funny, because I do think that’s an important divide between sort of where we all started together.
Andrew Radin: Yeah, no, I would, I would agree with that. I mean, look, most of our focus, these days is not on discovery. It is actually in the development of the therapeutics. It is about, you know, preparing for IND filings. It’s all the regulatory work we need to do there. It’s medicinal chemistry. It’s a whole bunch of things that are outside of the discovery process. And as we proceed to the clinic, more and more of our overall effort as an organization has less to do about the core innovation that created all of these assets and more about the heavy lifting you have to do to ultimately get that product to market.
And I think, to kind of tie it back to my previous comments, I think there’s been a new generation of capabilities that has been created. To what these guys just said, it’s gonna be a while until we actually see those things in the clinic. And to Tim’s point about, you know, computer vision and radiology, like there’s, there’s a lot of good science that’s already there and has been shown, experimentally to do a better job than obviously the, the human looking at those images. But yeah, it it’s gonna take awhile until that becomes the standard. I am, you know, my daughter was born almost five years ago now, but I was shocked to observe, even back then, which is only five years ago, that medical records were being passed from clinic to clinic with a fax machine. It just blew my mind. Like you gotta be kidding me, a fax machine? I don’t think I’ve seen a fax machine in all these years.
And so, yeah, I think part of it is, if you want to take the place where innovation moves the slowest it’s certainly got to be, you know, government, healthcare, or education. I’m not sure which of those might be the slowest, but there is a time for these new technologies to permeate the industry. And that is going to take time. And I think that’s when, ultimately, patients and the people that are on the receiving end of all this innovation, like that’s, when they’re going to see that difference. And it is going to take many years for this stuff to kind of make its way through the process and ultimately into the hands of providers and ultimately to patients. And that big benefit is going to come in the years to come. It’s obviously not in front of patients in many cases.
Harry Glorikian: Yeah, well, maybe my brain is wired towards risk or innovation because I’m like, “well, if you’re, if you wait till it’s done to get involved, you’re way too late,” right. You’re going to be a dinosaur or you’re going to be obsolete. And we’ve seen that in a lot of areas of tech compared to, you know, old standard industry.
There was a great piece the other day about this engineer at Ford who had been working on the gas engine for 40 years and then wakes up one morning and he’s like, I need to take early retirement because software and electric EV is the way it’s going to go. And now I’m just in this sort of maintenance mode of what I’m doing.
And I think about healthcare and I’m like any institution that isn’t at least dabbling in using image analytics. for radiology or something and starting to get used to this, I think they’re way behind where they may want to be in the next five years, because technology doesn’t follow just a slow curve on the way up. It has a way to go straight up at one point it before moving into an exponential curve. And I think the same for you guys. I mean, those companies that are not involved are partnering, investing in entities like you guys is, if you wait till it’s finished, you’re, it’s already too late. Because Andrew, your system will keep kicking out new molecules and Peyton, you’ll be making new antibodies and it’ll be a little too late to catch up. I mean, that’s, that’s the way I think about it.
Andrew Radin: I would temper that a little bit and the reason I would say that is because the companies that have been successful in the past in creating diagnostics and therapeutics…Products are on patent. They have long life cycles and they generate lots and lots of cash. And so, you know, big pharma, big diagnostics companies, they can kind of wait around and sort of see how things shake out with different younger companies and simply, buy or acquire, assuming that the companies are willing to be acquired. And so I think, large firms have been very successful in becoming, you know, acquisition and essentially manufacturing and marketing machines.
So I don’t necessarily think that some of these larger and established players that they’re necessarily, their livelihoods are threatened. I think they will continue to acquire the best of the best with their, with their large cash reserves. I think some companies in this space will gather the momentum and break out. And I think in time we might see some changes over time as to what the big, you know, sort of players are in this space. But it’s unlike other industries. Certainly software. It’s like MySpace disappears and Facebook reappears the next day. And that’s because you can deploy new technology and move users over in the course of an afternoon. And from a therapeutic perspective or a diagnostics perspective, that’s just not that the pace at which those things move.
So there’s, there’s lots of room for that. You know, and maybe similar in the automotive industry, you kind of have to build a factory and build some cars. It takes some times, right? So, so maybe there’s some parallels there, I think in some cases, but. I don’t see like a wholesale change happening overnight. At least from where I stand.
Harry Glorikian: Not overnight, but we definitely have to have dinner and like have a discussion around this topic. Because I would love to bring some examples to the table about how I see things. Once you digitize something, the model itself doesn’t have to stay the same way as it used to be. It is up for change. So I think those are the shifts that may change the dynamics of the market.
But I’d love to have that discussion with a wonderful glass of wine. After having come from Napa this week, I can show up with a few nice bottles. Thank you so much for taking the time. Andrew, thank you for bringing this group together. Peyton, Tim, it was wonderful to meet both of you. I hope that we stay in touch and I’ll keep watching the companies as they, progress. And I wish you guys incredible success.
Peyton Greenside: Thanks so much.
Tim Sweeney: Thank you Harry.
Andrew Radin: It was our pleasure.
Tim Sweeney: Andrew, Peyton, good to see you as always.
Andrew Radin: Absolutely.
Peyton Greenside: You too.
Harry Glorikian: That’s it for this week’s show. You can find past episodes of MoneyBall Medicine at my website, glorikian.com, under the tab “Podcast.” And you can follow me on Twitter at hglorikian. Thanks for listening, and we’ll be back soon with our next interview.