Xaira Therapeutics debuted with $1 billion in committed capital in April 2024.
“It debuted with $1 billion in committed capital in April of 2024.”
The Long Run with Luke Timmerman
12 Nov 2025 · 1h 12m · Luke Timmerman, Marc Tessier-Lavigne · 27 quoted facts
Marc Tessier-Lavigne argues that drug discovery is still a craft built on intuition and trial and error, and that AI can turn it into something closer to engineering. His company, Xaira Therapeutics, is attacking all three steps at once: picking a target, designing the drug, and matching it to the right patients. He says the goal is to cut development time in half and double or triple clinical success rates.
The central tension is near-term proof versus long-term bet. Designing antibodies with AI is already working and is aimed at hard-to-drug targets. Building "causal" models of cells, which would show what drives disease, is a multi-year effort that needs huge amounts of new experimental data. Xaira therefore runs as both a platform and a product company, using drug programs to keep the AI honest.
Using generative AI building on David Baker's work, the company makes millions of designs and tests hundreds of thousands in the wet lab. Results feed back into the model. Tessier-Lavigne's image is designing the needle rather than searching the haystack.
Targets like VEGF and EGFR are already well served by existing antibody methods. Xaira starts with hard targets, such as multi-pass membrane proteins that barely poke out of the cell, because AI aims at a chosen spot instead of relying on screening.
Models trained on descriptive data, like single-cell RNA data, can sort cell types but cannot say what change turns a diseased cell healthy. Tessier-Lavigne says they have never seen causal data, so Xaira generates gene-knockout data (Perturb-seq) at scale.
Xaira published what it says is the largest Perturb-seq dataset to date, which drew 64,000 downloads in three months plus inbound interest from companies. He argues the field needs many labs contributing, while each keeps some proprietary data.
He cites about 13 years from target to approval and roughly a 10% clinical success rate today. By 2035 he wants about six and a half years and success rates of 20 to 40 percent, with AI as one major contributor among several methods.
Xaira is a South San Francisco company, launched with $1 billion in committed capital, that uses AI across three steps of drug discovery: choosing a target, designing the drug, and finding the patients most likely to respond. It also builds its own drug programs to prove the technology works.
Traditionally, researchers screen huge libraries hoping to find a molecule that binds a target. Tessier-Lavigne says generative AI instead designs the antibody or protein from scratch aimed at a specific spot on the target, which helps with hard targets that screening handles poorly.
It means the model can predict what happens if you change something in a cell, for example which edits would turn a diseased cell back to a healthy one. Today's models only describe differences between cells, so Xaira generates experiments in which genes are switched off one at a time.
Tessier-Lavigne says the field needs data from many cell types and many labs, and that Xaira wanted to contribute to that effort. The release drew 64,000 downloads in three months and inbound interest from companies, though Xaira keeps some data proprietary.
He gives no specific timeline, only that Xaira has launched therapeutic programs and wants to reach the clinic quickly on the right targets. He describes the broader causal-biology work as a two-to-five-year effort.
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
Xaira Therapeutics debuted with $1 billion in committed capital in April 2024.
“It debuted with $1 billion in committed capital in April of 2024.”
Xaira Therapeutics was co-founded and incubated by Arch Venture Partners and Foresight Labs.
“Xaira was co-founded and incubated by Arch Venture Partners and Foresight Labs.”
David Baker won the 2024 Nobel Prize for work on computational protein design.
“Designing protein drugs on computers with ideal properties is a big idea that won Baker the Nobel Prize in 2024.”
Marc Tessier-Lavigne's father, a Canadian military officer, was posted to London and then Brussels with the NATO delegation.
“he was eventually sent to Europe, to London, and then Brussels with the NATO delegation out there.”
Marc Tessier-Lavigne went to Oxford on a Rhodes Scholarship after studying physics at McGill.
“I was fortunate to get a scholarship, a Rhodes Scholarship, to go to Oxford.”
At Oxford, Marc Tessier-Lavigne earned a second undergraduate degree in physiology and philosophy, where the physiology covered neuroscience and biochemistry.
“So I did a degree in physiology and philosophy, and the physiology was neuroscience and biochemistry.”
Marc Tessier-Lavigne did his PhD at University College London under David Attwell, as Attwell's first PhD student.
“I followed David to do my PhD with him as his first PhD student.”
Marc Tessier-Lavigne took his first faculty position, at UCSF, in 1991.
“So I took my first faculty position in 1991.”
Tessier-Lavigne's lab identified some of the first chemoattractant factors that guide axons in the brain, which it named netrins.
“we were fortunate to identify some of the first chemotractants in the brain, which we called netrons”
Early in his lab's work, Tessier-Lavigne was funded in part by the Paralyzed Veterans of America spinal cord research foundation.
“one of my funding sources was the paralyzed veterans of america spinal cord research foundation”
When he joined Genentech in 2003, Tessier-Lavigne oversaw about two-thirds of the research portfolio, including oncology.
“I was brought in to oversee about two-thirds of the research portfolio, including oncology”
Genentech, which had spun out its earlier neuroscience work, rebuilt a neurodegeneration research group during Tessier-Lavigne's tenure.
“So we actually built back up a neuro group at Genentech.”
Denali Therapeutics was co-founded in 2015 by Marc Tessier-Lavigne with Ryan Watts and Alex Schuth, encouraged by Bob Nelson.
“Bob encouraged me as well as two colleagues from Genentech, Ryan Watts and Alex Shuth, to co-found Denali in 2015.”
After leaving Genentech, Tessier-Lavigne joined the boards of Pfizer and Regeneron.
“I also joined two larger companies, the boards of Pfizer and Regeneron.”
During Tessier-Lavigne's presidency, Stanford University opened its first new school in 75 years, the School of Sustainability.
“including opening Stanford's first new school in 75 years, the School of Sustainability”
After leaving Stanford, Marc Tessier-Lavigne was contacted by Bob Nelson about a company idea with Vic Bajaj and David Baker that became Xaira.
“I pretty rapidly got a call from none other than Bob Nelson, who told me about this exciting idea that he and Vic Bajaj and David Baker were working on”
Tessier-Lavigne describes drug discovery as choosing a target, making a drug that acts on it, and identifying patients likely to respond.
“if you think about the drug discovery process, you have to choose a target. You have to make the drug that has the intended effect on the target.”
David Baker's lab published RFdiffusion in 2023 and RFantibody in 2024, generative AI tools for designing proteins and antibodies.
“David's advances with generative AI with RF diffusion in 2023 and RF antibody in 2024.”
Protein-design models are trained in large part on the curated structures in the Protein Data Bank.
“The data that forms the foundation of that is the highly curated data in the protein data bank, the PED.”
Xaira says it makes millions of AI designs and tests hundreds of thousands of them in the lab, feeding results back to its models.
“making millions of designs, testing hundreds of thousands of them in the wet lab in these virtual cycles”
Xaira is first applying AI antibody design to hard-to-drug targets such as multi-pass membrane proteins, which have little surface exposed outside the cell.
“Think multi-pass membrane proteins with small amounts of real estate on the outside of the cell”
scGPT, an early single-cell foundation model, was created by Bo Wang, who now works at Xaira.
“single cell GPT, SCGPT, created by Bo Lang, who's now here at Xaira, for example, very powerful”
Perturb-seq inactivates one gene at a time and measures the effect on all roughly 20,000 genes in the cell.
“one gene at a time, you modify the gene, you inactivate the gene, look at the effect on all 20 ,000 genes in the cell.”
In June, Xaira published a methods paper and what it described as the largest Perturb-seq datasets to date, from two cell lines.
“We just, back in June, published the largest PerturbSeq datasets to date”
Xaira's released Perturb-seq data was downloaded about 64,000 times in three months.
“there have been, I think now, 64 ,000 downloads of the data in just three months.”
Xaira has a sizable site in Seattle near David Baker's lab, alongside its South San Francisco headquarters and a small London group.
“We have a sizable site in Seattle near in proximity to David Baker's lab.”
Tessier-Lavigne cites about 13 years on average from target to FDA approval, with only about one in ten clinical drugs succeeding.
“it's 13 years on average to go from having a target, making the drug and getting it through clinical trials to FDA approval”
The host frames Xaira as a platform company combining protein design with large biological datasets, backed by $1 billion.
“building a platform for drug discovery that gathers multiple types of biological data”
A French-schooled upbringing and a switch from physics to physiology and philosophy at Oxford set up his scientific career.
“Oxford gave me my vocation as a neuroscientist”
His lab's work on axon guidance, and meeting spinal-cord-injury patients, pushed him toward applying science in industry.
“it isn't just an intellectual pursuit, it's something that can help people”
He explains why a neuroscientist joined a company with no neuroscience program and what he learned there about leadership.
“go somewhere where I had a chance of being successful in the near term”
He addresses the research-integrity episode and describes how a call from Bob Nelson led him to Xaira.
“stood by the quality and integrity of my body of research as a whole”
He lays out three steps of discovery and argues AI can replace artisanal trial and error with engineering.
“we can transform this artisanal process into something that's closer to an engineering discipline”
Generative design lets Xaira aim at specific spots on proteins that screening struggles to hit.
“we're trying to design the needle”
Existing cell models describe but cannot predict interventions, so Xaira is generating perturbation data at scale.
“the model doesn't know what it doesn't know. It's never seen causal data.”
He explains why Xaira released Perturb-seq data and how project teams combine biologists and AI scientists across three sites.
“it'll take a village and it will take multiple labs generating data”
He sets aspirational goals for shorter timelines and doubled or tripled clinical success rates.
“we should aim to cut the timeline in half”
Welcome to the Long Run. This is a podcast for biotech adventurers. I'm your host, Luke Timmerman. Today's guest is Marc Tessier-Lavigne. Mark is the chairman and CEO of South San Francisco-based Xaira Therapeutics. Xaira is using computers to design protein drugs from scratch, building on work from co-founder David Baker's lab at the University of Washington. Designing protein drugs on computers with ideal properties is a big idea that won Baker the Nobel Prize in 2024.
But there's more to the company. Xaira is building a platform for drug discovery that gathers multiple types of biological data genomic, proteomic, etc. and feeds that into AI models to train them and hopefully tease apart what's causing disease versus what's merely hanging along for the ride. That information on what's causing disease is supposed to help guide the design of these new targeted protein drugs from the ground up. The hope at Xaira and all other AI-assisted drug discovery companies is ultimately to create a system that can make drug R &D faster, less expensive, and more predictable.
Xaira was co-founded and incubated by Arch Venture Partners and Foresight Labs. It debuted with $1 billion in committed capital in April of 2024. Mark is a neuroscientist. He was on the faculty at UCSF before he took an R &D leadership role at Genentech during its run as the leading cancer drug developer of the early 2000s. He returned to academia as president at Rockefeller University and then later Stanford University. He left Stanford two years ago amidst some controversy before coming back to industry with Xaira in 2024.
Right now, he has the advantage of being in a startup at a moment of technological possibility. Xaira has a lot of money, brilliant people, and a clean slate to do something consequential for all of BioPharma R &D. It doesn't need to unlearn a lot of things to start doing them a different and better way. And before we get started, a word from the sponsor of the long run, Dash Bio. Are you tired of inconsistent bioanalysis results and waiting months for data that should take days dash is the only bioanalysis cro built from the ground up with a tech first approach designed to deliver better faster and cheaper than anyone else with dash you get faster turnaround with results in days
not months high quality data across major assay types including eliza msd lcms and pcr supporting all modalities and therapeutic areas customer first policies like guaranteed outcomes and transparent pricing. From preclinical to late stage studies, DASH helps you move from assay development and validation to sample analysis with unmatched speed. Founded by industry veterans who felt the pain of traditional CROs, DASH is the partner researchers and clinical leaders actually need. Reliable, fast, and easy to work with. So if slow bioanalysis CROs are costing you money and miss deadlines, put DASH to the test.
Visit www.dash.bio to see how fast bioanalysis can be. That's www.dash.bio. Please enjoy this conversation now with Marc Tessier-Lavigne on The Long Run. Marc Tessier-Lavigne, welcome to The Long Run. Thank you very much, Luke. Thank you for having me. So, Mark, there's a lot I want to cover here about your life in biotech and at your new company. But first off, how do I pronounce the name of this company?
It's Xaira. Xaira Therapeutics, X-A -I-R -A. And we like the alliteration. Runs with Thera. Does anybody ever mess that up? All the time. um zaira zaira uh Xaira um it turns out that that it it's a it's a thing um uh we we posted little tiktok style videos um on our linkedin page and made us one of the most trafficked uh linkedin pages uh at the time uh while people were making fun of the name so what was a bug is now a feature
okay all right well you know as entrepreneurs do you you make uh lemonade right Right. Okay. Well, before we get into what you're doing there at Xaira, I'd like to ask a little bit about your background. So I know you're Canadian. It's Trenton, Ontario, I think was your hometown. What was that place like?
it's a a tiny town uh i was actually born in trenton it's a military base um my father was in um the military his his entire career in fact my mother was as well they met in the military um and so i was born on a base we actually lived in some towns nearby belleville and uh brighton you know on lake ontario so small town but i was a small kid um so i had a fine childhood there And this is how far away from the major city of Toronto?
It's about halfway between Toronto and Kingston, if you know, sort of on either end of Lake Ontario, so several hours away. Okay, okay. So what did your parents do in the military? My father went through a number of jobs. He started as a pilot, but then got into various kinds of management positions. In fact, he was eventually sent to Europe, to London, and then Brussels with the NATO delegation out there. I should say my mother was also in the Canadian Armed Forces, the Air Force, like my father, that's where they met.
After they married and my older sibling was born, she retired from service. But we were very much a military family.
okay and what uh age were you when you moved over to europe i was seven we moved to london
i was there for a couple of years and then uh then moved to brussels uh obviously following my my family and i was there through high school um in in when we were in the uk and in belgium they my parents put us in the french schools the french lices you know they have lices around the world There's one in New York, there's one in San Francisco, because I thought that wherever he ended up, there would probably be one of those and there'd be continuity in my education.
And it was for me, it was wonderful to I started out in English, but then all my education was in French. And that was a real gift to have that that opportunity.
OK, OK. So say a little bit more about these schools. What kind of schools did you attend and did you move around from one to the other? that's how we think of the military life in the United States, that military kids move around a
lot and have to fit into different environments. Yeah. Again, the great thing about my parents' decision to put us in this school system is that there was continuity. So even when we moved from London to Brussels, I could just pick up the curriculum. It was the same as in London or other places. It was a very traditional French curriculum. We prepared for the baccalaureate, a lot of focus on growth learning, but also certainly for me, some of the high points, the opportunity to really go deep in math and in physics, which is what ended up being what I was most interested in, but also having broad exposure to lots of other areas too.
So, you said math and physics were a couple of your favorite subjects. What else interested you as a kid, I guess, in middle school, high school, coming of age?
Yeah, well, in terms of subjects at school, I was actually, interestingly, although I ended up being a biologist, I was not that interested in biology. There, it was really just very rote, learning a lot of facts, not a lot of analytical thinking. I guess I liked anything that involved problem solving. So the other subject that I really was intrigued by was philosophy. which, again, in the French schooling system, perhaps less so in some other schooling systems, was a staple in high school.
So I'd say the subjects involved solving problems are ones that really appeal to me. With always thinking that I, at that time, I thought that I'd become a mathematician when I grew up.
Interesting. Well, you mentioned the rote learning a couple of times. And that's often how people first get exposed to biology and why a lot of kids find it kind of dull. And it sounds like, I mean, you felt that way too, at least for a while.
Absolutely. I thought it wasn't that it was an interesting, it was interesting to know these facts, but it's not something that really appealed to me or got me going. It wasn't until much later that I realized that there's as much problem-solving and analytical thinking that goes into biology as into other subjects. But that was for later. So, yeah, at the time, did I find it dull? No, it was modestly interesting, as interesting as some other subjects that also involved mostly learning about facts.
Were you always a good student or did it take you a while to kind of hit your stride?
I was, you know, I was a pretty studious student. You know, I think, yeah, I was pretty applied to my studies. Not that they were the only thing. I had lots of friends and I enjoyed lots of other outdoor activities, but I was pretty focused on my studies. and partly also because I actually derived a lot of joy from math in particular and physics as well. So just that problem solving is something that I love doing.
I'd say to this day, one of the things I love most is math. I don't have the opportunity to do it very often, but just pure, unadulterated joy. Interesting.
Okay. So I know you went back to Canada. You got a physics degree at McGill. You went back to Europe again for Oxford for graduate school, I believe.
To do a second undergraduate degree. Actually, that was an important transition for me. So, you know, when I got to college, I was very single-minded. I took as many physics and math classes as I could. But I started with both, but then I ended up focusing more on physics. And I was fortunate to get a scholarship, a Rhodes Scholarship, to go to Oxford. And I signed up for a PhD in physics. I had a thesis advisor selected.
But then something interesting happened to me. I just feel, again, just very fortunate. In those days, the Rhodes Scholars would assemble in New York from North America, Canada and the U.S. and take the QE2 over to England, you know, five days together. And the people I met on that trip were just so extraordinary, so broad, and none of them really planned to focus at Oxford. They all saw this as an opportunity to broaden their horizons.
And so I thought, gosh, why rush into a PhD in physics? I can go back to that and hear the beauty of Oxford and the scholarship is that I could switch at the last minute. And I did a second undergraduate degree. And there I had by then become, my interest had been piqued in biology because I did my undergraduate thesis in a biophysics lab. And they had this program that at some level doesn't make a lot of sense called physiology and philosophy.
And I've been intrigued by philosophy. So I did a degree in physiology and philosophy, and the physiology was neuroscience and biochemistry. And that's where I just fell in love with the nervous system. Oxford gave me my vocation as a neuroscientist, and philosophy there really taught me how to think. It was a very important part of learning and growing. So I'm just so grateful for that opportunity. Without that, I'd probably be a physicist today. I'd probably be very happy doing that, but I'm so grateful that I had that opportunity.
It sounds like you were really bridging those worlds between science and humanities, philosophy and physiology. These two don't usually go together. I don't think in very many programs.
Well, it is actually, it's curious. I think a lot of people who know the Oxford system know there's a famous degree called PPE, Philosophy, Politics, and Economics, and they all tie together. It's a go-to for anybody interested in political science or government or things like that. And they had this other program called PPP, Philosophy, Psychology, and Physiology. And you can see how those tie together, those three. But in the second and third years, which as a senior student, I was allowed to skip the first year, you only choose two of the three.
And I was probably the first person in five years, I think it was, to choose that particular combination, which doesn't make a lot of sense. Most people would choose Physiology and Psychology. Those fit together nicely. or psychology and philosophy, which fit together nicely. So there was a logic to it, but I just was very lucky that there was something that was just perfect for me.
Now, who would have been some of your favorite philosophers to read back then? And then maybe who would have been an inspiring figure in physiology or the beginnings of neuroscience?
Well, what I got in Oxford was a heavy dose of British analytical philosophy, which I loved. I'd grown up as a child with continental philosophy, great world systems, but the analytical approach was much more in tune with, I guess, the way my brain works. So whether it's philosophers like Strawson, who made Kant intelligible to me, some of the more contemporary British analytical philosophers like Gareth Evans, who just passed away when I got to Oxford, Christopher Peacock, who was my teacher and extraordinary mind.
I just learned so much from him. That's the school that really got my brain working and then reinterpreting a lot of prior philosophy through that analytical lens. In terms of physiology, I was very fortunate having as a tutor Julian Jack. Any electrophysiologist who's listening to this will have used Jack Noble and Chen, Chen being Dick Chen, that I think a lot of people know, Dennis Noble. So Julian was actually a very lively intellect from New Zealand, but there in Oxford.
And he was a great mentor because he directed me towards some really excellent other tutors who were there, including the person I ended up doing my PhD with, David Atwell, who at the time was a fellow there and then moved to University College London. And I followed David to do my PhD with him as his first PhD student. Okay, okay.
So what was it about biology that really captivated you and made you decide that you actually want to become a research biologist?
Well, I guess the evolution for me was when I was a physics undergrad. I just loved math and physics. a lot of the action was occurring in high energy physics. And here, it wasn't really a scientific decision. It was more a personal, almost sociological decision. I didn't want to be part of a thousand-person team living in an accelerator or doing things like that. And even though I liked the problems, I was more interested in a smaller setting and also where I could go back and forth with experiment.
And so that's how I ended up focusing on condensed matter physics and why I signed up for a PhD in condensed matter physics. It was almost sociological. But when I did my undergraduate thesis in biophysics, I realized, gosh, this is sort of similar to what appealed to me about condensed matter physics, uh except hey let's face it more interesting because it's about life um and so that's sort of how i it was sort of the this understanding what i liked about physics if you will as a profession as opposed to just an a fun exercise or something to learn about uh and then when i went to oxford that's why i studied those two subjects and um i the nervous
system just uh understanding how the brain works just seemed like such a fascinating problem. And being able to take that kind of approach also appealed to me very much, and that's how I ended up going up in that direction. Interestingly, when I started my PhD, I focused on precisely that, the processing of signals and noise in the retina, some mathematical modeling, some experimental activity. And in the course of doing those experiments, I just became more and more fascinated, actually, with the complexity of the brain and the question of how it gets wired up.
And so that was my next evolution during my PhD too. I finished my PhD working on brain function, if you will, but then moved to do post-doctoral work. I wanted to work on brain development, how the brain gets wired up. And so I was very focused on choosing a lab where I could actually go tackle that problem. And also by then I realized I needed to learn molecular biology, which was really, it was the molecular biology revolution in those days.
And I was fortunate to be able to go do postdoctoral work at Columbia in a lab that was really bringing together molecular biology together with studies of development.
Now, what years are you talking about here in graduate school and your postdoc?
So graduate school was in the mid-80s and postdoc was the late 80s and early 90s. So I took my first faculty position in 1991.
one okay okay but going into neuroscience i mean yes the questions are so big about how the brain works and the technologies the tools for evaluating what's happening are really were were and are pretty limited um so how did you where did you start uh in terms of thinking about okay how how can I make some headway and gain some some understanding with in in this uh just vast
domain the in terms of brain function there I I part of um the reason I I joined the advisor did was uh precisely because he'd chosen a system where you could tackle some of those issues uh the retina is complex but relatively simple compared to a lot of the rest of the brain and And it was actually possible to measure, you know, his program was to measure the properties of individual cells, their connections, and build models for the flow of information as light hits the photoreceptors and goes to the next layer and the next layer, and how it gets processed down a lot of contrast enhancement that's occurring there.
So that seemed like a piece of the nervous system that was tractable. And I made some contributions as part of my PhD to understanding that process, some of the filtering of noise that occurs at nonlinear synapses, for example. But it was pretty clear also that it would be a long road to really having sharp enough tools to be able to go at it in a big way, which, by the way, we are there today with with some of the tools that have been developed, like optogenetic tools by my colleague, and others.
That came later, right? And this is even before functional MRI.
Correct. So this was early days. But as I said, in the course of doing that work, I developed another fascination, which is with brain development, brain wiring. So I guess we'd be having a different conversation if I'd stuck with studies of brain function. But for me, this was just a way to get introduced to what ended up being my passion for my postdoctoral work. And in fact, the rest of my scientific career, it's always been a major focus of mine to focus on how the brain gets wired up during embryonic development.
So these are basic questions. Were you thinking that this was going to be your path forever? Like faculty, basic research into neuroscience, how the brain works, how it's wired?
Absolutely. At that stage in my life, as a postdoc and then as a starting assistant professor, I took my first faculty position at UCSF. That brought me to the West Coast. I was single-mindedly focused on trying to understand how is it that nerve cells, neurons, make connections with the appropriate target cells to form the neural circuits that underlie all of the brain functions. sensation, control of movement, cognition, and so forth. It's just extraordinary, you know, 100 billion neurons in the brain, all wired up perfectly to have just the right circuits for brain function.
And as an entry into that, what I believed early on was that the problem would give in the area of axon guidance. The neurons, when they're born, send out a thin cable-like extension, the axon tipped by a motile structure or the growth cone that sort of leads the growing axon through the thicket of other nerve cells to find just the right targets. And when I started working on this, it was known that axons are guided by chemical cues, that there are attractive factors that the axons will turn towards as they grow and extend, and then there are repulsive factors that they'll grow away from, but none of the molecules were known.
And I really believe that this was a problem that was ripe for, to tackle. And so my ambition and my only ambition in those postdoctoral and early faculty days was to try to identify the factors that cause the brain to wire up.
So you came out to the West Coast, San Francisco, early nineties, and you're starting up a lab and you're pursuing this research agenda, which is pretty broad, some pretty big questions. But then, of course, I mean, you must have met some people in biotech. What did you make of this nascent biotech industry and how you might work with these folks? Well, there were two steps.
The first step was that in my lab, we set out to identify the brain wiring factors, and And we were fortunate to identify some of the first chemotractants in the brain, which we called netrons, and then built on that and also set up some close collaborations with other scientists to really try to flesh out what's the full complement of guidance cues in the brain. So this was sort of a realization of that ambition that we went from not knowing any of the factors to having a pretty good idea of how it works.
And at that point, I guess I peeked up my head for two reasons. One was I became interested in the application of my work, and that was caused by two events. um one i uh when i set up my lab i my ambition was to identify these factors i i didn't apply to the nih it was clear the nih would be too risky a project so i'd look for other funding sources and one of my funding sources was the paralyzed veterans of america spinal cord research foundation and they provided money for me to go after these elusive factors and there was one requirement which is every year i had to go to the annual meeting and and the first time
I went to the meeting, I met the members of the board of directors and they were all in wheelchairs. I'd, of course, in my grants explained how these factors might be helpful in rewiring the nervous system after spinal cord injuries, which is why I applied to the Granting Foundation, but it was more an intellectual point when I met them and they were just incredibly supportive and generous, you know, saying, hey, we know that this, it'll take time to translate discoveries like this, but we want to support you to do that.
I realized it just became very personal. And I realized, boy, it isn't just an intellectual pursuit, it's something that can help people. And around the same time, my father had a stroke, and I saw the devastation that comes with the injury to nerve cells when axons are damaged, when nerve cells die, and the the brain is disconnected from the spinal cord, leading to paralysis that can often be permanent. And those two events made me want to apply my work.
And I started doing some translational work in my lab. And then being in the Valley, exactly what you said, I was fortunate through a colleague to be brought onto the scientific advisory board of a startup that was just starting, Exalexis, which is still going strong today. And one of the co-founders, Corey Goodman, is a good friend and we were close colleagues. So I joined as an advisor and I was just astonished by what could be accomplished in the private sector.
So it was that combination of wanting to see the application of my work together with an exposure to biotech that got me thinking, Five years later, Corey and I actually, and a third colleague, Antito Serafini, co-founded a company, Renovus. But during that period, I realized that if I really wanted to go to the distance, I'd have to go to the private sector myself. And that made me a sitting duck when, in the early 2000s, I was approached by Genentech to join them.
Okay, well, we'll get to that in a second. But I think this is fascinating how you saw those connections between your academic research and working on these basic factors. And if you could identify them and characterize them and publish this work and advance the field, there would be somebody in industry who might be able to develop compounds that could interact with them. And perhaps, you know, with a little bit of luck and time and money and effort, help people in wheelchairs to walk, to regain ability to walk, or stroke victims to be able to move their arms and legs on one side.
That's very tangible.
Yeah, that's right. It was the human impact and the ability to have an impact to help people. To take this science, which I loved so much initially as an academic pursuit, but to realize that by pursuing science, you have the opportunity to help people in need. That was really a very profound motivator that's really carried me through to this day.
So Genentech calls in the early 2000s. And I mean, at that time, I mean, they were primarily, I mean, they were on their road to becoming the world's biggest cancer drug maker and antibodies, antibodies. They were the best at this. What did they want with you, a neuroscientist?
That's a great idea. In fact, they had, so this was 2003. And in fact, they just spun out neuroscience into a company called Renat. and so I was being recruited to a biotech company, a neuroscientist being recruited to a company that wasn't focused on neuroscience anymore and so the question is, why were they interested in me and why was I interested in them? I mean, they'll have to answer the first question but I can tell you from my perspective, precisely because I'd worked with colleagues and co-founding a neurology startup, what I learned was just how difficult it is to make progress on the nervous system.
Even to this day, it's one of the big frontiers in biopharma, right? It's just harder to tackle the problems in the nervous system. And I realized if I was going to go into industry, I'd rather go somewhere where I had a chance of being successful in the near term. The problem with tackling a problem where you aren't successful is if you're not successful, you don't know whether it's because the problem was too difficult or because you weren't very competent.
And I wanted to have a problem, you know, situation where I had some chance of success so I could gauge how well I was doing. And that made it actually, was actually appealing to me that Genentech did not have a neuroscience program. I was brought in to oversee about two-thirds of the research portfolio, including oncology, which, as you pointed out, by then was probably one of the largest efforts in the industry. Certainly, Genentech became the number one oncology company in the world during that period, during the 2000s.
um and uh in fact interestingly a few years later as the company continued to grow and had a lot of success with just some uh amazing innovative medicines that were approved one after another um in the the um the early 2000s as you'll recall um the uh we we had the opportunity first to to double down on oncology and and immunology um but um but also to ask should we go into any other areas. And so we evaluated this and decided to go into neurodegeneration.
So we actually built back up a neuro group at Genentech. So it was a bit of a homecoming for me with neuroscience, but in a very data-driven way, based on where the science was and evaluating other opportunities for the company as well. So that's really where I learned a number of things. One about drug discovery, of course, where I had everything, I had to learn everything, but I was learning from the best, but also about executive leadership.
Just an extraordinary place, you know, from the CEO, Art Levinson, on down through the executive committee. Just extraordinary leaders from whom I learned so much on a day-to-day basis.
Yeah, I think that would have been when we first talked and I got to know a lot of those amazing executive leaders, some that you mentioned. And so you're there, I think, through the Roche acquisition. Then you made a move back to academia. We can kind of skip through this pretty quickly. You went to Rockefeller, I believe, and then later to Stanford. But you stayed in touch with your industry contacts, like serving on SABs. What were you doing during those years immediately post-Genentech?
Yeah, so I was recruited to Rockefeller University. And so, you know, Rockefeller is amazing. I mean, just an extraordinary place. And then I was also blessed to be recruited to Stanford five years later. In terms of what you're alluding to there, which is how I remained involved with biotech, after I left Genentech, I was approached by a number of companies and I joined the boards of Agios. there I met Bob Nelson when he co-founded Juno.
He brought me onto that board as well. And then Bob encouraged me as well as two colleagues from Genentech, Ryan Watts and Alex Shuth, to co-found Denali in 2015. So I stayed in touch with small biotech, if you will, but It was another very broadening experience, just amazing companies. All three, Ryan and Alex, of course, went into Denali and they and their colleagues have made it what it is today. I also joined two larger companies, the boards of Pfizer and Regeneron.
So I was fortunate during that entire period to still have a front row seat to a lot of the great innovation that was happening in the industry.
Are you tired of inconsistent bioanalysis results and waiting months for data that should take days? Dash is the only bioanalysis CRO built from the ground up with a tech-first approach, designed to deliver better, faster, and cheaper than anyone else. With Dash, you get faster turnaround with results in days, not months. High-quality data across major assay types, including ELISA MSD, LC-MS, and PCR, supporting all modalities and therapeutic areas. and customer-first policies, like guaranteed outcomes and transparent pricing.
From preclinical to late-stage studies, DASH helps you move from assay development and validation to sample analysis with unmatched speed. Founded by industry veterans who felt the pain of traditional CROs, DASH is the partner researchers and clinical leaders actually need. Reliable, fast, and easy to work with. So if slow bioanalysis CROs are costing you money and miss deadlines, put DASH to the test. Visit www.dash.bio and see how fast bioanalysis can be. That's www.dash.bio.
So you're a busy guy. You're wired in with some of the most interesting projects going on in cancer and in neuroscience in those years. You go to become president at Stanford. And of course, you were there for several years. And then there was a big story that happened about an investigation into some past articles, some research papers that were not reproducible. Some allegations of research misconduct were made. And I mean, I didn't follow it that closely at the time, but as I understand, the Stanford board did it or they conducted an investigation and found that you were not, they exonerated you, that you were not manipulating data, but that maybe there was some data that was not correct or in
those papers that they cited. and that perhaps there could have been some better oversight. I guess, you know, you ended up leaving. And what, I guess with a little bit of benefit of distance now, a couple years away from this, I'm sure it wasn't pleasant at the time. Are there any lessons that you take away from that experience?
Well, you know, maybe a few comments, Luke. First, I'm really proud of my time at Stanford and the groundbreaking work that we did together, including opening Stanford's first new school in 75 years, the School of Sustainability, transforming the university with some forward-looking initiatives in health, in learning, in social issues, in AI. and maybe one of the things I'm proudest of, which is setting up a core curriculum for first-year students that's focused on civics and that's engaging across political differences.
With respect to my research, you know, I've always stood by the quality and integrity of my body of research as a whole and panel that looked into issues that surfaced in a few of my papers, validated my scientific integrity. And with that in the past, I'm excited and focused entirely on the future, which is Xaira and our mission to transform drug discovery and development
through AI. You must have been hurt. What did it feel like when this happened? Because, I mean, president of Stanford, it's a high-profile job and it doesn't feel good to be featured in the New York Times and elsewhere, people making accusations about your integrity.
You know, Luke, the important thing, I think, in a situation like this is to go through the process, trust the process. And as you say, I was exonerated. And then to really look forward to the future. So not an easy time, but definitely, you know, created new opportunities for me. I I wouldn't be CEO of Xero right now at this extraordinary time if I were still in that position.
Okay. So what happened, I guess, in the immediate aftermath of this incident? Did you get a call from Bob Nelson or somebody else? Or did you take some time off to go to a remote island somewhere, clear your head?
Well, definitely took some time with the family. And of course, to refocus on my lab, I'd maintained a small research lab at Stanford during that period. But yes, I pretty rapidly got a call from none other than Bob Nelson, who told me about this exciting idea that he and Vic Bajaj and David Baker were working on to create a company that would apply AI to help transform how we discover and develop drugs end-to-end. And I started talking with them.
I joined them, providing advice, became a consultant, and pretty rapidly decided to go into the company. So it was very organic and just an exceptional opportunity. You know, this is a combination of a few things, I'd say. It's the moment. um ai we've all seen just the transformative effect that ai is having on so many business processes i believe that it's going to have a transformative effect on drug discovery as well um it's the people you know bob vick uh david uh david baker will be known to everyone who's listening to the this podcast of course the nobel prize a year ago um and uh um and also the vision the vision for the company so it was really the
the moment the people and and the vision and that got me very excited and made me go for it.
Well, yeah, these are often the ingredients of a big story. Prominent people, grand ambition, and a lot of money too. The headline number was $1 billion of venture capital committed to a startup company, which is still a lot of money. Um, for, for those, um, who maybe haven't, didn't follow the first batch of stories about Xaira, how would you describe what's, what's the big idea here? Maybe what's, what's different at, at this company compared with, um, some of the other AI startups people might've heard of.
Right.
So the, the, um, the idea behind the company is, uh, again, to rather than tackling any one particular segment of the drug discovery process, stepping back and looking at the three steps of drug discovery and tackling them in a very staged and strategic way, but believing that there are important synergies that can come from that kind of approach. So if you think about the drug discovery process, you have to choose a target. You have to make the drug that has the intended effect on the target.
And then, of course, you have to identify the patients who are most likely to respond to the drug. And each of those steps currently is very artisanal. A lot of science, of course, but also a lot of art, a lot of experience, intuition, empiricism, trial and error. And that's why there's so much attrition, why the timelines are so long, why the expense is so great. At the highest level, the promise of AI is that if we can harness enough data of the right kinds, that we can transform this artisanal process into something that's closer to an engineering discipline, where a lot more of the discovery can be done in silico with higher success rates and lower attrition and shorter timelines.
That's the top-line vision, and we decided with Xaira to tackle this broadly. At the same time, the best test of whether your models are successful is to see whether they can be applied to generate medicines that can make a difference to patients. So we're also a product company. So we're applying the technology to develop a pipeline of differentiated therapeutics. And it's a combination of platform on the one hand, applying AI to tackle the three steps of drug discovery and product that I think is a little bit unusual in this space and why we raised the capital that we did.
We believe there's a lot of synergy between pipeline to platform and products. to improve your AI models, you have to generate a lot of data, which we do in the context of developing specific products. And also by applying the technology to making products, we're making sure that we're developing AI that actually is meaningful. If you're not tied to products, you can often go off in directions that end up being less useful for the intended purpose, in this case, to make drugs.
So it was that vision of tackling AI broadly, being both a platform company and a product company and having the resources to be able to push this forward and take some risks that drove us to the capital structure that we have.
Well, I want to ask you in a bit about the products because all platform companies eventually need to generate products, product candidates, and they get evaluated on that. But in the early days, first step seems to be building the platform. And you mentioned the types of data. That's very important that you feed to the AI so that it can learn. Can you talk a little bit about what are the key ingredients of this platform? How does it work?
What types of data are you feeding it?
So let's take the three steps and we'll start with the middle one, which is making drugs. So there, we're building on the great advances by David Baker, using generative AI to design proteins and antibodies rather than screen for proteins and antibodies. You know, there's a nice metaphor, which I first heard from Sean McLean at AppSci, that traditional ways of getting drugs, you're looking for the needle in the haystack with generative AI, we're trying to design the needle.
You design the antibody from scratch with the properties that you think it should have.
Correct. And so for that, the kinds of data, first of all, this whole field, first of protein folding, of course, with alpha fold, and then the David's advances with generative AI with RF diffusion in 2023 and RF antibody in 2024. It's easy to forget that it's just two years ago that RF diffusion was first published. the first real ability to get high success rates in generating entirely novel proteins from scratch. The data that forms the foundation of that is the highly curated data in the protein data bank, the PED.
And there's a lot of information there that models can be trained on to go a certain distance. We're supplementing that with other data to make the models even more powerful in terms of their ability to generate. So we've been generating, and we go through with AI, of course, these design make test loops where your model will make a number of designs. We make them in the wet lab and we test them and then feed the information back to the model.
So we're generating at very large scale, making millions of designs, testing hundreds of thousands of them in the wet lab in these virtual cycles, even as we continue to improve our algorithms and try to squeeze as much information as possible out of the structures that are already there out in the public domain. So that's a whole set of data. We're fortunate to have some data in the public domain. There's a lot of data we're generating for another set of applications in the case of target ID.
There, maybe I can explain the approach that we're taking in the first instance, which is to try to develop causal models of biology.
Can I just pause for a second and make sure I understand what you said here about the drug? So this would be maybe taking a well-known target, like something like VEGF or EGFR, where you've got a good structure of the target and you want to develop different or try out different types of antibody shapes, structures, proper characteristics in silico. And then test it again in the wet lab to see, you know, is it delivering what you think?
Correct. So that would be the testing that we're doing. And actually your point here enables me to in turn make another point. I think many of your listeners will be thinking, well, gosh, there are lots of good ways of making antibodies already, but you need AI for. And the truth is, there are a lot of targets that are well served by existing methods. The ones that you just mentioned, right now, it's not clear that you need AI to get antibodies to that.
So where we're first applying the technology is to targets that are not as well served by current methods. Think multi-pass membrane proteins with small amounts of real estate on the outside of the cell, where often it's difficult to screen for antibodies. Often you get few and not hitting the epitopes that you're interested in. AI is not limited by the difficulty of screening because the AI is always targeted at a particular epitope, particular spot on the target protein.
So the AI doesn't care that it's just, you know, the tip of the iceberg, it's peeking outside of the cell. So those are the use cases that we're going for initially. But we are training on lots of different kinds of data, including those from proteins that are more easily tested in the lab. But it's important in any company like ours, as your listeners know, to differentiate from the competition. And the way we're differentiating is by focusing on hard-to-drug targets, the so-called undruggable antibody targets in the first instance.
On the surface of cells or inside cells too?
Well, with antibodies, we're going for surface proteins right now. That's the near-term promise. getting hits against hard-to-drug targets, and then we can turn those hits into leads, optimize leads, development candidates. The beauty of AI, though, is that it learns, and so we can infuse other properties into the model. For example, train it on data for developability, working on reducing immunogenicity, increasing affinity. So over time, we believe we we will be able to make our models more and more powerful.
So with the ambition that when we start a program, rather than getting a hit that has to be sequentially turned into the various steps to a development candidate, that we can more rapidly get something closer to a lead or an optimized lead, or even perhaps close to a development candidate. We believe that time advantage will come. And obviously our ambition is to get there as rapidly as possible. We can get started with drugging the undruggable, But over time, we believe that AI will enable us to also gain a time advantage.
Do you throw lots of different types of structures at these hard targets like, you know, monoclonals, bispecifics, antibody drug conjugates, any and all of the above?
Well, we look at it in a modular point of view. So we're looking at the, you know, the binding regions of antibodies. We are looking at various formats, you know, VHHs, SCFVs, full-length IgGs. And then you can take those pieces and assemble them into bispecifics or multispecifics. Okay.
Now, you also mentioned target discovery. Yes. What sorts of data? How are you tackling that? Yeah.
So the way we're tackling that, and this is a longer-term endeavor, just to be clear, we see near-term value creation through applying AI to making antibodies. This is, you know, over, we're looking at a three to five year kind of journey where we believe that what will be helpful with target identification or just as important, you know, in some cases we know a lot of targets and the target prioritization or matching of targets to patients, which is really where the big prize comes, right?
that what we need are models of biology that are causal in the sense that they understand cells and then, you know, higher order structures at a causal level. The kind of question we'd like to be able to ask is we have a cell in a disease state. We'd like to bring it back to a healthy state. What changes need to be made in the cell to bring it back to the healthy state? Now today, you can with, we have very powerful tools for saying, oh, here's the cell in the disease state, here's the cell in the healthy state.
There are 300 genes that are different between them. And actually, some of the first AI models right on the heels of chat GPT back in 2022, rapidly, you know, various foundation models were created that, you know, GPT, single cell GPT, SCGPT, created by Bo Lang, who's now here at Xaira, for example, very powerful. they take descriptive data sets, single-cell RNA-seq data sets, for example, and those models can make sense of them. They're very good at descriptive tasks like delineating what kinds of cells you have in your mixture or identifying the differences between different cell types.
What those models fail at, however, pretty much uniformly is causal questions. The one I asked, if you have a cell and disease state, what do you have to do to bring it back to a healthy state? And the reason we believe at the core the model can't predict is because the model doesn't know what it doesn't know. It's never seen causal data. So we and a number of other companies and academics are busy trying to generate or generating large amounts of causal data where we perturb cells and look at the outcome.
There are various different ways of perturbing cells. One of the foundational ways that we're using is so-called perturb-seq, where one gene at a time, you modify the gene, you inactivate the gene, look at the effect on all 20 ,000 genes in the cell. So it's a 20 ,000 gene perturbation individually by 20 ,000 gene readout, the kind of data sets that you need to train models. and then doing this across a lot of different cell types.
We just, back in June, published the largest PerturbSeq datasets to date, and we are generating a lot more of those data. And those data, we believe, will provide us with a first model that will begin to understand cells in a causal way. um the the to go the distance though we need to um take uh that kind of scaffolding from cells that are relatively easy to to study which involves a lot of cell lines certain primary cells that we can do this uh genome scale perturbation on and others including some in vivo perturbations but then you need to to with with these models a lot of the power comes from fine tuning on data that are harder to get that
to come by, like diseased patient data with diseased cells and healthy cells. And so the theory here is that a foundation model of biology that is based on large amounts of perturbation data will help the model understand at a causal level the relationships within the cell and can begin to provide insights on tissues that are harder to come by from disease.
What you really want are these causal perturbations. If you know what the driver mutation is, for instance, and not just a passenger, you can aim your drug at the driver. And you have to do all of this heavy lifting to just go systematically down the line with genes in, say, neurons or heart cells or blood cells of different kinds. This sounds like a lot of data that you're going to have to collect and then feed to the AI before it can really start making those distinctions between what's a driver and what's a passenger.
That is correct. And unfortunately, we're generating data like that. There are big efforts in the academic world, for example, at CZI, the Chan Zuckerberg Initiative, the ARC Institute, and others that are also generating data like that. And, you know, the, I think.
They're talking about a virtual cell and you are too. So that you would have so much data that you can just run a simulated model of like what will happen if you had a drug aimed at this target in silico. Before you ever like actually synthesize it and run it through animals, that has to happen, but that happens later. You could really increase the odds of success and shorten the timeframes. like you were alluding to earlier, if you had a much better idea of whether this drug candidate's
going to fly. Absolutely. And that's really the aspiration, I think, for the virtual cell initiative as a whole. And it is a grand experiment as we look to see just how much data do you need and what different kinds of data. I've talked about ProTurtSeq, There are others who are doing important work with drug perturbations, for example, of cells to build this kind of causal model. And how much can you learn by protruding one cell type when it comes to another cell type?
And there the belief is that we're going to have to have quite a large compendium of data. And as I said, also then fine tune on data that are, I guess, more directly therapeutically relevant as well. So it's an exciting time. And as I said, we're making the investment. And we believe that this is something that will pay dividends over a longer time frame. We're looking at a two, three, four, five-year time frame.
So you've been there a year and a half or so, I guess. You've hired a bunch of really senior people. You've established some locations, got this pipeline or the platform up and going. What else can you say about your progress to date? And when might we expect to see some product candidates emerge?
Well, we're not giving specific timelines, but I can give you a general sense. We came out of stealth about a year and a half ago. with some just amazing scientists, some of the co-founders from David Baker's group and also other scientists who joined from some other companies as well. We fleshed out the leadership team. We've launched our programs in molecular design where we made a huge progress compared to the models that are out there in the public domain right now for antibody generation.
but which gives us enough substrate, has given us enough substrate to get started on our therapeutic programs, but where we believe there's still a lot more progress that's possible and that we're driving towards. So we have launched therapeutic programs. We are eager to get into the clinic as rapidly as possible, but we want to get into the clinic on the right kinds of targets as well. A lot of work on that. So stay tuned on that front.
In terms of the foundation models of biology, as I mentioned in June, we published a methods paper where we've industrialized the whole PerturbSeq platform and also two data sets from two cell lines that we used really to establish the platform. We want to share those data. We thought they'd be useful to the community in general and help galvanize some of the activities that are out there. But we're generating much more data on that front. So there we've actually made public some of our findings.
Yeah. Why did you put the PerturbSeek data in the public domain? I mean, you could have kept that internal.
We thought that, first of all, we're going to have to generate an awful lot of data. And it's important. This is an area where we need, you know, it'll take a village and it will take multiple labs generating data. Each lab will have its own focus on some data sets that may be more proprietary. We certainly will. But we'll also need a scaffold of data from lots of different cell types. and we wanted to contribute to that general effort.
And so, yeah, we put up more Perturb Seek data than had been published by the academic world before that. And we're excited to see what others can contribute as well to help accelerate the field.
Did people come to you with some good ideas after you put that out there on BioArchive? Or maybe you were able to recruit some people who said, oh, I can see what you're doing now and maybe I can help?
Well, we discovered a couple of things. First of all, there have been, I think now, 64 ,000 downloads of the data in just three months. So it tells you that there's a lot of interest in these kinds of activities. That's been a little bit, as some people call it, testing the market. We've also been approached by a number of companies that have been interested as a result. So it did have that effect. We really just put the data out there because we thought it was important to do so, but it's been really gratifying to see the response we've had from both the academic and the private sector.
How do you manage a company like this? Like a startup, how many employees do you have at present? I think we have 148 employees. Okay. And you've got people in some different geographies like South San Francisco, Seattle, probably some remote people. you got people with different like very different skills in technology and biology what is your role as CEO of this thing yeah so the so first I'm very blessed I just have
extraordinary scientists here we have extraordinary leaders just a great executive team seasoned people from the tech world, from the biotech world, from pharma. We do have three sites. We have our major site where we do the drug discovery work, the high-throughput biology. A lot of the data generation is here in South San Francisco where I'm located. We have a sizable site in Seattle near in proximity to David Baker's lab. David, of course, is a co-founder with just some amazing AI scientists there, as well as a wet lab footprint, too.
We think it's important to have a rapid cycle between the AI design and the wet lab. And we have a small group in London also. And this is really driven by talent that we were able to bring together just an extraordinary group of talented scientists there. and one of whom had worked with David, and so there was that strong connection as well. So we do have multiple sites, we have multiple kinds of science going on, and one of the joys of this whole enterprise is bringing together with such diverse backgrounds, you know, biologists, drug hunters, AI scientists, different kinds of AI also, So, you know, the scientists are focused on molecular design, scientists focused on biomedical AI, and coming together
with this common purpose of transforming how we discover and develop drugs. In practice, what really, where the rubber hits the road is on the project teams, where scientists will be working on a particular therapeutic target, where you have biologists, drug hunters, and AI scientists all working together. And it's just so inspiring to them to see and learn from each other. The biologists are amazed at what the AI scientists can do. The AI scientists are just thrilled to learn about biology, to learn about drug discovery.
And so we really try to bond everybody together through this culture that's focused around projects. But that's what you have to do in a situation like this, in these interdisciplinary spaces and where you bring together interdisciplinary teams. The art is to figure out how to get people to work together productively. And we're doing it by having people work together on projects.
You can organize them around something really big and hard and exciting, like, say, drugging RAS or P53 or, you know, I could go down the list. like, wow, if anybody could do that, that would be something. But, you know, just kind of wrap this up, Mark, you know, you've been, I guess you joined Genentech 2003, so it's been 20 years or so in industry. I started writing about it around this similar time. And, you know, if we were to have the conversation then, we'd say, well, drug candidates entering the clinic have about a 10 % success rate in clinical trials.
And it generally takes at least 10 years and, you know, a billion or 2 billion or depending on how you do the math to develop a new drug. Not a lot has changed there, but this is where the promise is. What you're talking about here is to change how this R &D enterprise works. If you look out 10 years from now, if you can dare, to 2035, what do you think this R &D, this biotech drug R &D enterprise is going to look like in terms of success rates, time, and expense?
So I like the 10-year timeframe. And I think that we should aspire. I really believe in the power of AI to help turbocharge each of those three steps in drug discovery. We haven't even talked about what's already happening with AI in the clinic, with identifying sites, recruiting patients, writing protocols, and so forth, where people, just dramatic efficiencies that are being driven there. So there's lots of opportunities to apply AI to improve efficiencies, in addition to what we're focused on, which is applying AI to deepen the science and, you know, increase success rates.
I think our aspiration should be, you know, currently it takes what, Luke, you know, I think it's 13 years on average to go from having a target, making the drug and getting it through clinical trials to FDA approval. And that it's about, you know, nine out of 10 drugs, only one out of 10 drugs succeeds in the clinic. I think as an industry and using AI, but also all the other methods that people are applying, because this isn't for AI alone, but certainly with AI, I believe as a major contributor, we should aim to cut the timeline in half.
We have to go from target to FDA-approved drug in six and a half years. Some will be longer, some will be shorter. that requires being able to go from target to in the clinic in just a few years. That's what we should be the challenge ourselves to do while simultaneously increasing the chances of success. And there, the biggest one, of course, is in the clinic, the massive failure rate and doubling, tripling the success rates from the very low levels that we have right now.
One in 10, 10 % getting to 20%, to 30%, 40%. That should be our aspiration over the next decade. And I can't see why if we apply ourselves well with AI, but also with all the other insights that people are bringing using other approaches that we can't get there. That has to be our aspiration. What we're doing today, just not sustainable, as we know. For sure.
But if you were able to actually achieve those kind of gains, I mean, that could really usher in a new era of abundant medicines. I mean, for lots of people who really have limited options today, we could have a much, much bigger medical toolkit 10, 20 years out.
Absolutely. And that's what we should aspire to. This is ultimately about applying the science in the best way to help patients as rapidly as possible. There's still so much unmet medical need. It's extraordinary to see the advances that we've witnessed over the past 30 years, 20 years, even 10 years. I mean, some of the breathtaking advances even the last five years. But we know that there's so much more that we need to do. And for that, we just have to be more successful than we've been in the past.
And I believe that AI is going to be a major part of that transformation.
I hope to be around a long time and see this era come to pass. This will happen in our lifetime.
I believe so, too.
Marc Tessier-Lavigne, thank you so much for joining me today on The Long Run.
Luke, thank you so much. I really enjoyed it.
Thanks for listening to The Long Run, a production of Timmerman Report. Pedro Rosado of Hitstepper Media was the sound editor. Music is from D.A. Wallach. See you next episode.