Martin Borch Jensen is a scientist, entrepreneur and longevity advocate. He is the co-founder and CSO of Gordian Biotechnology, a company whose platform enables the simultaneous delivery and testing of hundreds of therapeutics in individual animals. They have raised over $60M from top investors like Founders Fund, Horizons Ventures, Fifty Years and the Longevity Fund. They also recently announced a partnership with Pfizer to apply Gordian’s proprietary mosaic screening platform to accelerate the discovery of therapeutic targets for obesity.
Martin is also a prominent voice in the longevity community and activist. As founder and president of Norn Group, a do tank for longevity, he launched the Impetus Grants program, which has deployed roughly $34 million to scientists across 145 projects, with funding decisions made within 3 weeks to enable speed.
Watch on YouTube. Listen on Spotify or Apple Podcasts.
Chapter Markers
0:00 Intro
03:27 Where is the longevity field today and are we on track to cure aging in our lifetimes?
14:25 Is longevity truly different from other areas of biotechnology?
16:07 What is aging?
22:13 An aging body is like a company that’s developed toxic bureacracy
26:46 Gordian’s approach to tackling aging
45:09 Will scale alone solve biology?
49:08 What data would superintelligent AI need to cure aging?
54:00 Unified theory of aging or biology
59:26 What understudied areas in aging deserve more attention?
1:02:23 How does loss of cellular identity relate to aging and disease?
1:10:07 Why are there no trillion-dollar biotechs and what would it take to create one?
1:17:32 What are the key components needed for a biotech flywheel?
1:21:43 How can biotech companies de-risk clinical development?
1:37:02 What is Norn Group and what problems does it address?
1:43:58 What opportunities exist for individuals to impact the aging field?
Transcript
Daniel 00:03:17
Martin Borch Jensen, thank you for joining us on the podcast.
Martin 00:03:21
Thanks for doing this. We need more quality stuff.
Daniel 00:03:23
Awesome. Well, quality stuff, I don’t know, but stuff we can provide.
03:27 Where is the longevity field today and are we on track to cure aging in our lifetimes?
Daniel 00:03:29
I want to start by understanding where is the longevity field today? Are we on track to cure aging in our lifetimes?
Martin 00:03:38
No. That’s the default state. How would we cure aging? There’s at least two ways of thinking about it. One of them is there will be a magical one silver bullet, and that’s going to just solve all of aging. I think that’s usually the way it works in movies. It’s unlikely, in my opinion, to be the way it works in biology because as someone who’s been in the aging field for 15 years, we don’t have one aging. That’s one of the risks or pitfalls for people getting into the field. You think of aging as one thing, where it’s really a system of things that’s not behaving appropriately.
Silver bullet, maybe, but I think no. What is our other option? View aging as a series of problems, a series of things where the human physiology gets misaligned over time for a variety of causes, and we have to fix them faster than they go wrong. A way that you could quantify that would be what is our life expectancy and how is that changing? Today, from biomedicine broadly, with the longevity field contributing minimally, we’re getting something like 0.1 year per year. You have a curve that goes up of how long we’re living, and we’re getting 0.1 year per year. If we had more than 1 year per year, then maybe you could say that we’ve cured aging, at least if you’re able to sustain that for a long period of time.
We’re a tenth of the way there in terms of our rate of progress. The longevity field is not contributing very much to that. You couldn’t point to a drug that got approved that was discreetly a longevity drug that made a big impact there. You can take GLP-1s and then post hoc say these are longevity drugs because look at how much stuff they affect. And that may be accurate as well. But that would be our best bet. Otherwise, the additional years of life are just coming from we made a better heart drug, we have better sanitation, we have better vaccines, et cetera, et cetera.
That’s where we are right now. Then you could break down when will the aging field start having an impact. Here, of course, we have to recognize that in one sense, the aging field is thousands of years old. Go back to Gilgamesh, go back to first emperor of China. He was like, I want to live forever. And so he hired a bunch of longevity consultants to tell him what to do. And they’re like, you should drink mercury. It’s got divine power or something. And then he died. And then they buried him with the Terracotta Warriors.
In that sense, it’s been going on for a very long time. But in terms of we can do a manipulation and this actually changes the rate of aging in some animal, there’s two arguably first points in time. One of them would be the early Roy Walford caloric restriction stuff. In that case, we’ve maybe had a field for 70 years. Or you could say, good, but then there was a big lag up until 1990, 1991 when Tom Johnson and Cynthia Kenyon both showed that you could mutate worms, single gene, and then they will live for twice as long.
That’s like 35 years ago. I think that’s a better modern, we’re actually studying this deliberately, but tiny field at the time. Still tiny field, but even tinier, make fun of you sort of stage of the field.
Daniel 00:07:29
Nobody’s getting made fun of anymore.
Martin 00:07:31
I mean, they should be, people selling supplements. But anyway, that’s 35 years ago. How long does it normally take some new scientific idea to get translated into now this is helping humans? Of course, there’s a wide range, but for most things in the realm of medicines, you’ve got initial science stuff, maybe like a couple of additional papers backing it up or different labs replicating it. Then you’ve got a company turns this into—you have an idea, a therapeutic hypothesis, then you need a company to turn that into a physical instantiation of the idea, like with a pill or whatever, but something that can actually go perturb the real world. That typically takes 5 to 10 years. And for aging stuff, anything that’s super new, it’s more on the high side. That’s a decade.
And then you go do clinical trials. And aside from the fact that for aging, we don’t know exactly what trial to run and what would get reimbursed in the end, and that’s a whole topic, bookmark that one. It’ll probably take at least 10 years. That’s like what it takes for slow progressive diseases of aging like Alzheimer’s. Once you’re through phase 1, 2, 3, you’ve got another decade. So consider a 20-year lag.
Only stuff that we came up with in the first 15 years of the field, at which point there were probably, I don’t know, a couple hundred researchers at most, might be less, doing a few things. Only that stuff has had time to actually become—and that ignores how likely is it that we pursue it as a company? Is the overturn window ready? Do we know how to design the clinical trials and so forth?
In that regard, you could say, well, maybe we are now at the point where stuff from 2005-ish could have been becoming like the actual therapies. And maybe in the next 10 years, there’s a lot of stuff that we discovered in that period of time that was fairly productive for the size of the field. That’s where we have things like rapamycin extending mice lifespan robustly with late age treatment. That was 2009, I think, Harrison. The first parabiosis papers, Irina Kornboyev’s, was 2005. Senescent cell stuff was around 2010.
There’s a bunch of stuff there that might be in trials now. And I think Karl Flieger did a nice overview of stuff that’s in trials, and that could pan out to be longevity drugs. So that may increase our rate of gaining years of life. If we’re at 0.1%, maybe the whole longevity field has been maturing from like just like 4 people to a few hundred people. It’s probably a couple of thousand people now. It’s hard to estimate. And some real things were done. And some things that were not so real were done and some early companies sold based on that. And we’re going to start adding some number to that.
How would we estimate the number? What would I guess if I had to make some Polymarket bet on longevity drugs that are labeled or clearly biologically are targeting longevity mechanisms, how much are they adding to our life over the next decade? One way you could think about that is in my nonprofit Norn Group that I think we’ll talk about, we did a rough estimate. We had like a dozen people on a call working in aging biology. How many new ideas could we quantify? We just made a whole list and then we divided by number of years. And I’m sure this is very rough, it could be off by threefold in either direction, but it was like 2.5 per year or something. New ideas in the sense of if we target this protein, then that could help lifespan or healthy years. Or if we do partial reprogramming, that could be a thing. Or if we kill senescent cells. So it’s fairly broad.
But okay, if that’s like 2 or 3. In general, when you start a biotech company and you want to make a drug, 99% of things fail. 90% of things fail before you get to clinical trials, and then 90% of things fail in clinical
Daniel 00:12:05
Eric and I, I sent him a text last night. I saw, I think from phase 1 to going to market, oncology drugs fail like 90-something percent of the time.
Eric 00:12:17
Oncology is particularly difficult as a disease area.
Daniel 00:12:20
But that’s phase 1 to market, right? That’s not even initial idea. That’s horrific.
Martin 00:12:27
It’s something like 99%, and it depends on which area, but let’s just assume aging is not on the easier side. You’ve got 3 ideas per year, 99% of them fail. Then how much are they going to yield? I think it would be fairly generous, but not crazy, to say maybe 2 years of healthy life if one of these things pans out. Then how many of the ideas do we actually turn into companies? Right now, that’s fairly high because there’s a low number of initial ideas and there’s a reasonable amount of excitement for aging therapies. So maybe it’s 50%.
If you add those numbers up, you get 0.03 successes at 50%. So let’s call it 2 instead for easier math. You’ve got 0.02 ideas, you try half of them. So 0.01, and then you get 2 years if it succeeds. Then maybe the longevity field could be expected currently to be adding 0.02 years of life per year, which on the one hand is really cool and validating the thesis that aging drives a lot of diseases. Because if you can add 20% to what all of biomedicine is doing with less than 1% of the budget, then you’re doing very well. On the other hand, you’re nowhere near 1, right? We’re not on track to treat aging, and you need to change some of those numbers in the system in order to be on track.
Eric 00:14:05
I’m curious to, this is an area of hot debate, but is longevity truly different from other areas of biotechnology and drug discovery? Because in a sense, all drugs are healthspan and lifespan extending drugs by definition.
Martin 00:14:20
Except for chemo.
Eric 00:14:21
Except for chemo.
Martin 00:14:22
And some other things. I think part of that comes back to what is your definition of aging? What is this thing you are targeting? If you believe that there is one aging process and you could target it with a silver bullet, then it’s in some sense distinct from everything else. Everything else will be unable to improve that thing. I don’t think that’s true.
I tend to describe longevity medicine as evidence-based multimorbidity drugs that work. Evidence-based as in we know that aging is the number one risk factor for most of the diseases that are the biggest unmet needs and most of the things that are the biggest source of mortality in the United States and many other places. Where would you start? If you’re pretending that we were inventing medicine from scratch, you’d be like, oh, only the old people are dying. Maybe we should figure out why, right? That’s fairly clear and evidence-based.
Multimorbidity drugs, as in you are not targeting a narrowly defined disease and one thing that goes wrong in a symptomatic or particular idiosyncratic way. You are broadly improving health. I think that’s generally part of people’s definition of improving aging. And then of course it has to work. So in that sense, no, it’s not that different, but it is a place to start in the same way that you could say targeting metabolism or targeting inflammation would be a useful idea for how do we add more healthy years. The evidence is so compelling. Old people die.
14:25 Is longevity truly different from other areas of biotechnology?
16:07 What is aging?
Daniel 00:16:08
One of the hard things though about tackling aging is, like you said, it’s many things and people have a very hard time understanding what aging is. It’s not clear we really understand what it is yet versus other diseases we treat. There’s very clear mechanisms, very clear pathways, or at least the diseases we’ve been successful in treating.
Martin 00:16:27
That’s an important thing. Do we have very clear mechanisms of Alzheimer’s? We have something that was a core hypothesis for a long time, and we failed to treat the disease by targeting that.
Daniel 00:16:36
Right. And the argument would be that’s a disease of aging and that’s why we need an aging lens to treat it.
Martin 00:16:41
You could say that could be true. It could also be that that’s not true, and Alzheimer’s is a particular dysfunction of the human brain, and it could have multiple causes. One of the things that drives it is clearly aging. But understanding the disease could happen independently of understanding aging if you have, in your causal diagram, something that you could target. Let’s say that it’s related to the stimulation of inflammation by chronic infections, which happen more with age and with escalation of chronic inflammation with age. But basically, if you can get the microglia to chill out and stop driving neuronal loss, that may be sufficient to effectively treat Alzheimer’s, even though it’s just a subpart of everything that happens in aging.
Daniel 00:17:36
So there’s a framing of aging that just came up, and I always have trouble grasping it, which is aging is the number one risk factor for disease. As age increases, death rate increases. Maybe aging is contributing to one aspect of Alzheimer’s, but it’s maybe not the whole thing. But you could replace the word aging with time, right? Everything requires time to progress. If you just froze time, no disease would progress. How do you distinguish, like what do you think the aging process is?
Martin 00:18:07
Two questions there. How do you distinguish time from some process progressing? And of course, if you only have instances where the process progresses with time, like in every organism and every individual, then it’s very hard to tease them apart. The experiment you would do is you would find a way to accelerate the process and then see if that led to the disease sooner. Can we cause accelerated aging? But of course, you’re still in this circular Gordian knot where, well, what is aging? We don’t know. Part of it may be DNA damage. And so if we induce DNA damage, do we get neurodegeneration sooner? Yes, we do. But is that truly aging or is that just a part of it? So it is true that it’s hard to tease those things apart.
I think in general, and maybe this will come up again and again, go make a change to the system and see how it behaves is the best way that you get the answers to those things. You test causality through perturbation and you could do that in either direction. Obviously, if we could treat Alzheimer’s, then it’s clearly not just time because we’ve reversed it unless we think we’re inventing time machines. So I think that’s true for Alzheimer’s. It is a good sort
Martin 00:19:31
I think to refute that aging drives disease, you could look at the incidence rates, which increase exponentially with age. It’s not a linear increase. You could get an exponential increase if you had a more complex system, which obviously you do, with individual different fail modes. If any of them fail, then you have some feedback loop where you get cascading failure or a vicious cycle kind of thing.
The way that I think about aging is basically the human body is an organizational system. Think of it even as an organization, like a company if you’re a startup type listening to this. If you’re a politician, think of it as a country. But you have different parts that each have to function and function correctly in relationship with each other in order to achieve a certain desired state. For a company, you’re trying to have revenue. For a human being, you can argue about what the meaning of life is. Dawkins will say it’s make babies. Other people will say do interesting things. But the capability of an organism to do any of those is inhibited by aging.
So you have these different parts. There are different ways that this thing can fail. Imagine you’re in your company. One department gets unhappy. They hire the wrong person, the manager sucks. You have something equivalent to inflammation where they’re angry all the time. For a company or an organization, it’s easy to imagine that clearly that will spread. Signals will come out of that, work products will come out that are shitty, and that will spread and impact the entire organization. That’s how I tend to think about aging.
You have a sophisticated instantiation of molecules that creates life. It’s fragile. Most distributions of molecules do not create life. It has self-regulating processes that self-preserve and continue that state. As soon as any of them—there are many ways that things can go wrong. The whole thing is designed to fix things that go wrong, but there’s any number of ways that the responses to what goes wrong will then cascade in some way.
If you think of an aspect of aging and disease like fibrosis, where your liver or your lungs will start developing scar tissue, it’s a good system until it fails. It’s a system that says we need to regenerate when you did some nasty thing to your lungs, you smoke a cigarette, you do chemotherapy, and a bunch of your lung cells die. We’ve intentionally created processes that trigger more cells to divide, we should create new extracellular matrix. But then you get into a loop where you have continuous feedback that is misaligned to the state of the system. It’s like a thermostat that—the thermostat’s over here and the heater is over here. You just keep heating the room even though the thermostat is outside. You have some sort of dynamic like that.
You can envision the same thing in the way that organizations communicate within different teams. If you don’t have the right feedback loop that keeps the system aligned and going in the desired direction and getting back to that state, then it falls apart. It falls apart in a way where the different pieces affect each other, which is why you see some people die of heart failure and some people die of Alzheimer’s. There’s definitely a genetic component to that. But the heart of the Alzheimer’s patient is also old. You never die at 75 with a 25-year-old’s heart. It’s always spreading to some degree, but there is variety.
That’s kind of how I think about it. But as you alluded to earlier, that’s a complicated thing to deal with. Like how do you fix a company? You hire a great CEO and then they’ll do all the right stuff, and we don’t fully understand what they’ll do because otherwise we’d just do it.
22:13 An aging body is like a company that’s developed toxic bureacracy
Daniel 00:23:56
And the analogy to an organization or company is so apt because also it’s in the name. Organism. It is. It’s an organization. It’s this superorganism. It’s the symbiosis that forms between trillions of cells. Maybe it’s like a trillion cells. It’s not surprising that the organization breaks down eventually. It’s hard. How do you maintain everything functioning?
Martin 00:24:18
And there are—if you think of it again as a company, most companies will become more bureaucratic over time. Something will go wrong and you will establish a process to prevent it from going wrong. We can think back to the fibrosis. At some point, you’re just really bad because you’ve created too much scar tissue. How does a company become effective? It has happened. Typically there is a strong driver of goal-directedness. Think of Apple V2 when Steve Jobs came back and he’s just like, nope, nope, nope. We’re doing it this way. Some sort of hardcore founder mode. Obviously, that appeals to us here. We’re in San Francisco right now.
But there is something that is driving goal-directedness. And where do we see that in organisms? You see it during development. You’ve got one cell, one fertilized egg, and that has to create a whole body. It’s a miracle that it goes so right almost all of the time. That’s so wild. There are all these subproblems there. I’m not even a developmental biologist, but how do you know what’s front and back when you’re just one cell?
But you have this whole process that creates humans successfully over and over and over again. If we think about how could you treat aging, can you reactivate programs that already exist in our biology the right amount in the right places to restore? That’s a popular approach to treating aging right now. We’ve found that this partial reprogramming—put some transcription factors in that partly turn you towards an embryonic state. But don’t go too far. There’s some promising evidence. It also seems complicated. That was 9 years ago and we’re just still figuring it out. As I said in the beginning, for early science, maybe one decade to figure out what to do and start trials and then another decade for the trials.
But that way of thinking where biology has a lot of stuff going on, how can we tap the biology to do things that are desirable to us? Then we don’t have to understand the system perfectly in this Newtonian sense, but rather we can redirect it. Hire smart people and let them do smart stuff kind of approach to aging.
Daniel 00:26:43
And what does that mean in practice? Maybe we could talk now about what Gordian is working on, which involves genetic perturbations. Do you see that in the framework you just described or do you think that’s a different model?
26:46 Gordian’s approach to tackling aging
Martin 00:27:00
I think Gordian is a subpart of that, but you need an additional piece to it. How could we envision reestablishing goal-directedness, circulating proteins? How does your kidney know what to do? How does it know the state of your brain?
Daniel 00:27:18
I talk to it every morning and I say, keep doing what you’re doing.
Martin 00:27:21
Your pancreas sends emails. The body has pathways of communication. Those are really complicated because usually you have one protein and it does five things per tissue, but different things across all the different tissues. Understanding all the circulating parts and how you would intervene and which tissue will drive the other tissues—maybe the brain is really good because it is a central regulator, maybe the liver is really good because it secretes a lot of stuff—that’s something to figure out, which is not what Gordian is engaged with right now.
But if we want to take this approach and nudge the system back into the right state, that is what Gordian is doing. The company is named after the Gordian knot. Back in ancient Greece, they believed a lot in their oracles. Maybe we’re doing a full circle to our LLM oracles now. But basically, some oracle said for the town of Phrygia, the first person who drives into town on an ox cart should be made king. They believed that. The peasant Gordias comes in on an ox cart, and congrats, you’re king now. Surprisingly, he does a good job and the city thrives.
Two generations later, his grandson Midas—this was before the whole gold incident, before he started gilding everything around him—wanted to make a commemorative monument to his grandfather. They took the ox cart that he drove into town on, and they tied it to an olive tree. They tied it so well with this knot that was so complicated that nobody could untie it. That’s where the concept Gordian knot comes in. It’s an impossible problem.
Now we zoom forward a little bit to the not-yet-great Alexander the Great.
Daniel 00:29:24
Who I saw, by the way, is an advisor to your company.
Martin 00:29:26
He’s going east, trying to take over the world. The oracle Pythia at Delphi has said that whoever manages to untie this knot will go on to take over the world. He’s heading east and that is exactly his plan. He makes a quick stop to fulfill the prophecy. He gets there, tries to untie the knot. He cannot. He can’t figure it out, but he is relentlessly resourceful. He takes out his sword and cuts the knot. He says, “Ah, there’s no rule against this.” The knot is cut. And then he went on. Maybe the fact that he cheated is why he didn’t quite make it—he only got half of India.
But that’s the story of the Gordian knot and the concept of finding a way to solve the problem that doesn’t involve untangling all the complexity. When we started the company, I spent 15-ish years studying aging and I still didn’t know how it worked. The only thing I had gleaned successfully was that nobody else knew exactly either. What do we do then? I want to make things for humans in our lifetime.
That’s where we came up with: if we have the system and we don’t fully understand it, but we have it, we have aging happening in a living organism, can we go into that system and make changes, try something causal, try a perturbation to understand how the system works and what the result of this particular action would be?
That’s the technology that we invented. We call it mosaic screening, where we put different interventions into an animal that has progressively developed some disease. Complex diseases of aging are generally what we work on. We put a very low dose of gene therapy into the animal so that you get individual interventions inside of different cells. You don’t change the whole organ. The whole organ is still diseased, but you have these independent cellular experiments that are happening in the environment of disease. Whatever is important, even if you don’t understand it—the immune system might play an important role or metabolism—that is present when you’re doing your experiment.
Then you pull the cells out and do single-cell transcriptomics. You measure the expression of every gene in these cells, and you can compare that to various different reference points: an animal that was never diseased, a human that is diseased, or a human that has been treated with some partially effective drug. Now you can say which of these interventions actually had a beneficial effect.
What we do is basically create these atlases of what is the best way to treat osteoarthritis or cardiorenal disease. We can do all these perturbations and I really would like to know how many in vivo treatments have been tried by anyone for a given disease. Within pharma, you don’t have the numbers. I’m pretty sure that the curve for humanity went and surpassed everything that had been done in every pharma. I would bet money on that, but I can’t find the numbers. It’s speculation for now.
But to create the whole atlas of every target that maybe is interesting to drug and then look at what they all do. If we go back a bit to the number of ideas that we’re producing for how we treat different diseases of aging—on the one hand, we’re going disease by disease for now rather than focusing on just the age of the cells. That’s because you solve one difficult problem at a time. Don’t try to solve what is a clinical trial for aging while you are solving hard science. Start with diseases where there is a clear path to what you can do. Later, the same platform works exactly the same for aging.
But going disease by disease, now you can 100x the number of ideas that are tested per year. We could cover one disease in a year and just test everything that’s worth trying to drug and just go one by one. Then now you can find answers in this systematic way. It’s boring. Let’s assume there’s no silver bullet. Let’s assume that you don’t need to be a genius. I designed the whole thing so I don’t have to be smart. Just methodically find out what the answer is in the environment where you want the answer to be true. Then stack those up.
Now we can go into one organ and say, how do we get you on the right track? The complement to that is not what we’re doing in our company. Not many people are doing it—Vadim Gladyshev a little bit. But what is that communication system within the organism that ages? What are the signals that are going from one organ to another? I think that’s a really cool topic that’s being studied a little bit, but it’s complicated to do.
Daniel 00:34:33
But I suppose right now you’re saying Gordian is focused very much on what’s happening at the cellular level.
Martin 00:34:39
Well, to the cells in the diseased organ.
Martin 00:34:48
Exactly. Like cells in a dish, you could say your autophagy is inhibited. Maybe if we increase it, you’ll be better off. But is that really what’s going on in the diseased organ? We go back to fibrosis, which is one of the areas you can work on in vivo. The reason that the cells are doing X, like secreting too much collagen, is because of the signals that they are getting from other cells in that diseased environment, because of the feedback loops that are present in this context, and because of the changes to the cells with age.
If you just take some young cells in a dish, well, you can find a way to cure Alzheimer’s if 20-year-olds got Alzheimer’s, but they for the most part don’t. There’s something else that happens. You have all of those things and then you ask what is the right nudge to a cell? How do we make an employee perform well in a mediocre company? Maybe that will spread from there. I think the more successful things will, but that’s really the question. We have a broken state already. How do we get started at getting back? And then you can scale that.
Eric 00:35:54
When you think about some of the questions that need to be answered about the underlying mechanisms, one of the ones that comes to mind for me is this idea of standardization. How do you really establish a standard of what a healthy cell should look like and then compare single-cell data from a pooled screen that you’re running here? What is healthy supposed to be? What’s the baseline you’re tracking for?
Martin 00:36:16
The thing that makes that particularly hard is that you want a dynamic answer. There’s no formal notation of biology. Think of different scientists. You have a way to describe the system. The ways that we have to describe biological systems are either static or obviously wrong, like differential equations of flux and stuff. But we know that if you like for 12 hours, things would work differently 100% of the time.
But aging doesn’t happen in vitro. In the place where it’s easy to do measurements over time of everything going on in a cell, you don’t have aging. Aging does happen to organisms, but one, slowly, and two, you can’t track everything that’s going on. To start to get at those, you need to have some way of capturing the dynamics of the system, not just the static state of the system.
That’s where, when I think about the data that you generate, like how is Gordian’s dataset fundamentally different than many other things? Like the tabula muris sinus, single-cell transcriptomics, every organ of mice, same thing exists for human throughout age, different time points. That’s great. If we feed that to GPT-7, will we cure all disease? I think no. You’re scowling as well.
Eric 00:37:42
Don’t think so.
Martin 00:37:43
Why not? What is the stuff that’s needed to make Dario’s dream come true, that AI will solve everything? I want AI to solve everything. I don’t want to do this job if I don’t have to. I mean, it’s fun, but you don’t understand how the system works from a static snapshot.
It’s like, I want to understand human society. I’m an alien, fly in, hover over New York City, and I’m like, okay, what’s going on here? Let me download Google Maps. And now I’ll understand human society because I see the whole thing. You can see the whole of human society except with some mines and satellites, but most of it’s on the ground. You can see the whole thing. But you would not understand what’s going on. Why is the person there? Where are they going? What happens? Why is it so different that this normal looking person goes into a bank versus this group of 5 people with balaclavas going into a bank? What’s going on there? Maybe you could figure it out, but without the time resolution, I don’t think you can.
What we are doing is we are introducing perturbations and we are asking if the cell was in this state to begin with and then you did this thing, what would then happen? What would happen over time? What’s the new equilibrium? You can kind of map the manifold of cell behavior if you do it to enough different cells. Now I think you can potentially feed this kind of data through AIs and have them, using AIs colloquially for any kind of intelligence that is now incredibly cheap, whether that is an LLM or some other thing. Now you could envision inferring causal relationships. That is something that with unlimited intelligence you could do a lot at scale and now start to make hypotheses around why is this thing going wrong.
Because there are some things already in bio that work very well, like protein structure, where you train on that data. But do you treat this disease is way underspecified. What even is this disease? Is it one disease? We call it Alzheimer’s. Is that one thing? Maybe it’s 5 different things with different drivers. How do we solve this, I think, is limited. If the answer was in the literature, maybe you could get to it faster. But solving those kinds of problems, you need different kinds of data input, like in vivo perturbation at scale in disease contexts with multiple reference points, I think is one thing that you can do.
Daniel 00:40:19
I have something very specific I want to understand. You have some animal model, let’s say for fibrosis. You’ve done a perturbation, it’s been a few days or something, whatever the time frame is since you did the perturbation. Now you want to do your single-cell sequencing. I assume at that point you have to kill the animal and extract the organ.
Martin 00:40:37
Most of the time, yes. Sometimes you could do a biopsy.
Daniel 00:40:40
Okay. In the case where you can’t do a biopsy, how do you, like, I wonder, are you going to get weird data because you’re now sequencing cells that died? How do you manage that? I’m sure this is a problem in bio.
Martin 00:40:52
Generally speaking, the cell has a lag of responding to anything. What we’re measuring in our case is primarily the expression of genes. There’s a certain speed at which transcription happens. It’s very fast, much faster than we can conduct the experiment within the living animal and at that temperature.
But there’s different tricks we do. Take one like put the whole thing on ice. Now the transcription doesn’t happen very much. There are other tricks that we have, but too detailed and proprietary. Basically freeze the snapshot at that point in time and prevent the cell from responding to the dissociation that you’re doing. That is something that’s reasonably well understood. Not everyone’s equally good at it, but we are.
Daniel 00:41:45
And also you don’t want crosstalk between your different perturbations, right? You’re doing it at some low enough penetration level. Not a very high percentage of the cells are getting perturbations. I guess it must be very tricky to balance that with doing enough that you can actually see some phenotypic impact at the organism level.
Martin 00:42:07
No.
Martin 00:42:09
That’s not in that first scale. You have your cellular state transcriptomes, and that is what you are mapping all the physiological stuff to. It’s not just that you’re looking at the health of the cell, because that would be of more limited value. But if you have paired datasets where you know that this is a heart that beats more strongly or is more or less fibrotic, and you know how that manifests as a signature in the transcriptome, you don’t have to understand everything from the transcriptome. It’s a mapping thing. You can make predictions from one cell of what’s going to happen. Or you could make predictions around, is this going to be toxic? Is this going to increase cholesterol in the organism based on that limited understanding?
The first animal stays sick, but the cells don’t. Then you follow up and you show that if you do this thing to all the cells, it actually gets healthy. Maybe that only happens half the time, but you’ve gone from 500 things to here’s 10 we can test.
Daniel 00:43:14
When you’re checking the transcriptome of these cells, one of the things you’re doing is you’re comparing it to the transcriptome of a healthy cell and seeing if it moved closer to that?
Martin 00:43:21
Yeah, although here again, if we think of biology, that’s true. And you compare it to cells within the same organism that have received negative controls where you’re not really perturbing them. Maybe positive controls, maybe there are some genes that we know can benefit this disease from genetic studies or drugs that are approved. Those are some comparators. You could also look at animals that have gotten better. Now it depends on the disease. For some diseases, if you’re inducing a stress, you can lay off it and then measure.
Because there’s path dependence in biology. It’s not like you have here’s healthy and here’s disease and it’s a one-dimensional axis. And it’s just like going this way and I need it to go straight back. It’s a manifold where certain paths are feasible and some are not. And the full state of a diseased cell involves both things that are bad and things that are trying to combat the bad stuff. You don’t actually want everything to go away, which is again, part of where having a lot of data is really useful. You need all this data in order to figure out what’s going on.
Daniel 00:44:30
Transcriptomics feels like a huge force multiplier in the field. It feels very powerful. But at the same time, my understanding is transcriptome data is highly dynamic, changing constantly. How do you get past—is it potentially a lot of junk? And does that get solved through just scale, like getting tons and tons of it?
Martin 00:44:53
I don’t think it gets solved by scale as we were discussing earlier. You need structured data in the right way. Maybe if you want to understand causality, you want perturbations or time course things, but time course is relatively hard. You could do cool things with molecular recorders and stuff. But I’m on team not good enough for, we will just sequence more and then we will understand the whole thing even with AI.
What makes me believe that? We have been doing a lot of transcriptomics even before we had single cell. We had bulk. There’s so much public data. There are companies that have been formed decades ago to just understand biology from transcriptomic data like Numerate. So far has not succeeded, generally speaking. We’ve created single cell atlases. How much have we gone from a single cell atlas to a transcriptomic measure of disease? Do we even have that? I kind of assumed that we did when we started the company. I didn’t start out as computational, and I was just naively assuming that we’ve done so much single-cell measuring disease, that must be a thing we’ve done. We have not.
45:09 Will scale alone solve biology?
Eric 00:46:08
A big part of that is exactly this idea of path dependency in biology, which is that you cannot simply measure a system as is and expect then that data to be immediately interpretable. The interpretability and ultimately interoperability of biological data is highly context-dependent and path-dependent. And unless you have a very tight control over the original context and the path that a system takes, you have essentially a bunch of variables you can’t account for with the result you’re measuring. That’s really tough. It’s super nonlinear.
Martin 00:46:39
I don’t think that just scale will get us there. It could be that that’s wrong. And you just need more scale. That is possible. I don’t see anyone who would really benefit from making that argument and pursuing that strategy as opposed to the, let’s add perturbations into the strategy. I don’t know that there is much debate there of, will we just solve it with scale? But my take would be probably not.
And that’s for animal models. Sometimes the animal models are really good. And that’s part of what we—the point of Gordian, and you can do everything in one animal or a few animals, is that you can go into animals that have spontaneously developed the disease. Our osteoarthritis program starts with horses that have developed osteoarthritis over a number of years instead of like we decide how disease works, create a model of it, and then cure the model, which now you just have obviously an additional layer of risk of if your initial assumption of disease mechanism was wrong, which it probably is because you haven’t made any progress yet, then you have a problem.
But even so, I think that’s much better. But even so, it is not a patient, and patients are not all the same either.
Daniel 00:47:55
Let’s say it’s a few years from now, maybe more years, OpenAI unveils GPT-50, superintelligence. What data is that superintelligence going to need in order to cure aging? I imagine a lot of transcriptomic data with perturbations, good controls. What else is it going to need?
Martin 00:48:17
I think that is definitely one of them, but then also we should think about humans. Transcriptomic is one layer. Transcriptomic is nice in the sense that it gives you a view on the whole cell state. You’re not limited to just one pathway, and it is doable at scale. Proteomics correlates with transcriptomics around 50%. It’s not that transcriptomics is everything we need, but proteomics is not scale enough yet. You probably want multiple layers.
There’s a bunch of different answers, but the short version would be you need a map of these paths. If we say that there’s path-dependent transitions, you need to have not just your end states, but enough points in between on that manifold in order to map it out and try to understand how you can get from A to B.
What that means depends on the task that you are trying to do. In this case, we are trying to make the human physiological system function well. We’re trying to bring that back to high-functioning state or youthful. Your data has to map the appropriate layers of organization or abstraction. You could take just the RNA of a cell, you could take the physical structure of the cell that exists in an organ with other cells and that exists in a body. I don’t think you need to go to the sociological stuff, but maybe.
If you want to draw an inference at the body level, then you probably need datasets that span these. I don’t know how much you can jump. Maybe you can only jump one. Maybe you need paired data that has the state of the cell with the state of an organ. Maybe you could jump more than one layer at a time. But I think we should be thinking about those things. What are the A and B points that we want to transition between? What are the ways that we can map the manifold?
If you just want the brute force solution, just map every state that a cell could possibly be in. I don’t even know if it’s feasible. But if you think of the manifold, how do you map a manifold more efficiently? You could track over time. You can follow one thing over time. That’s hard. I would encourage people to develop more molecular recorders where when something happens in a cell, it gets stored in some permanent way, such as a DNA CRISPR array. This is a thing that a few people are doing.
You could add perturbations. That gives you not a course, but it does give you a direction. It gives you what changed from state A to state A plus this perturbation. Now you at least have a direction, and then you can gradient descent to some degree and map the manifold. And then you want to do that at the appropriate layers of perturbation.
49:08 What data would superintelligent AI need to cure aging?
Eric 00:51:31
Maybe we can oversimplify a little bit of what Gordian’s doing. You have a few different things that you’re measuring. Ultimately you’re trying to interpret trajectories and manifold shapes from these measurements. You have animal model systems that model some aspect of disease biology, human disease biology. You have single-cell sequencing and mostly focus on transcriptomics. And then you have the ability to have an array of many-to-one genetic perturbations that you’re delivering to various cells within particular systems.
Is that enough for us to really map the important aspects of the disease? Are there pieces of data or pieces of perturbation biology that are critically missing from this? How do you think about the state of the union?
Martin 00:52:23
Depends on your goal. If your goal is to understand the disease perfectly, then there’s for sure many things missing, and you’re seeing a snapshot. If your goal is to find effective interventions, find therapies, then the cell state doesn’t have to give you a full understanding of what is going on as long as it, in some sort of latent way, includes are we going in the right direction or not?
We’ve demonstrated that. We’ve applied this to find therapies that are new for osteoarthritis. I don’t know everything that happens in osteoarthritis. The state of the osteoarthritis atlas at Gordian is I don’t fully understand osteoarthritis, but I can cure old mice, horse cells, human explants. We have a platinum standard data package that we have arrived at by this approach. We’ve narrowed down what are all the different options for what you could do very strongly.
Daniel 00:53:32
With data, in theory, we can map that manifold. And then we can figure out the interventions to get from one state to another. I think a tempting idea would be, once we’ve gotten a certain amount of data, all of a sudden somebody very smart or an AI can recognize a pattern. And maybe they develop a theory of how aging as a whole works or how a certain disease works.
Do you think it’s likely that we’re going to come up with an integrated theory of aging or even a theory of biology? Can we get to the same type of mathematical formalism we have in physics, but in biology, through this approach?
54:00 Unified theory of aging or biology
Martin 00:54:14
I think the way that we start to get to effective theories, models that are meaningfully predictive, will overlap with understanding the latent desires of biology. We’ve learned how to make pluripotent stem cells, and that was not by understanding everything that happens in a cell and how exactly to put the pieces into the right state. There’s some amount of brute force testing, and then we found these levers together unlock this intrinsic encoded drive in a cell to do this thing. We took that and now we’ve turned those pluripotent stem cells into muscle cells and brain cells. I think finding those feedback loops, those encoded behaviors, we could do more than once within the realm of aging.
Let’s take some different things that happen within aged cells. One of the things that happens is that you have your DNA, and every cell has the same source code, but it’s running different programs off of that source code in order to become a skin cell or a neuron. That creates an identity of a cell. There’s many studies that show this identity of the cell is partially lost with age.
What is going on there? It could be working in different ways. It could be that you need an active repression of certain elements of the genome, like these viral elements, retrotransposons. You just get a little worse and then they get a little free. It finally breaks free and then you die or that cell dies. What is the dynamic of that thing? If you could have the system of that, then you can tie that to another system, which is how do cells go senescent? How does inflammation affect, what is the response to inflammation?
I could imagine a chain of these things where you have a fairly good model of each individual thing. Biology exists somewhere between chemistry and then physiology and then sociology. All the different disciplines, when there are more moving pieces, you get worse at having your theories be predictive because there’s more caveats. We live somewhere at the edge of the hard sciences currently. We’re better than economics, but not as good as physics or chemistry.
You can just interpolate from there. Laws of physics work really well. Sometimes they break if you’re in weird extreme environments. Sociology, I don’t know if it works, maybe sometimes. We can’t really predict. Just take the stock market. We can’t predict stock market. We can’t predict elections. Clearly we’re not that good at it. That’s just because it’s harder.
Map biology on somewhere in the middle and then imagine shifting it this way. We can create chemical molecules. We do that for drug discovery. We can create chemical molecules and we can have some rules and we can figure out, generally this stuff will happen. We can synthesize a thing that does this.
Then the question becomes, how far are we in our current understanding of biology from what we want to be able to accomplish? Instead of having it framed as now we just understand the whole thing and we understand it perfectly, science never understands anything perfectly. I don’t think that’s really what we need. We just need to be able to fix stuff faster than it breaks.
How good does our understanding have to be of that? If you’re trying to fix your car faster than it breaks, you don’t need to understand the different quarks inside of your iron. You need some understanding at the appropriate level.
Going back to our hypothetical, what could a superintelligence do and be useful? I think we need to think about what is the right layer of understanding for the different things that we would like to achieve. What are the limiting things for the organism’s effective survival? They will probably exist at multiple levels. What is the right point of intervention? What is the data that we could generate in order to understand at that level what is the right way to nudge things backwards?
I think that we can build up somewhat modularly, but with a specific intention of understanding. Then we can try things and then some things will be dead ends and some won’t. But we should make sure we are at least trying the things that seem like they should be important.
In the longevity field right now, there are some topics that are very popular. Partial reprogramming is really popular right now. Epigenetic clocks is really popular. Nutrient sensing has been popular for decades and remains. There are other things that are not very popular, even though they seem clearly important. For example, the interplay between chronic infections and aging.
There’s all these individual things. Maybe you’ve heard about the strong correlation between mouth bacteria and neurodegeneration. Then recently there are some papers about the importance of Epstein-Barr virus.
59:26 What understudied areas in aging deserve more attention?
Daniel 01:00:07
Virus.
Martin 01:00:08
Exactly. There’s a strong thing here. There’s an interesting book from Paul something. I think it’s called The Coming Plague, and it’s basically saying we somehow drew a distinction between infectious diseases and chronic diseases. That was just how it is. Clearly there’s some truth there that there are some acute things that happen with infections, but also clearly papillomavirus causes cervical cancer, just unambiguously. There are cases where they overlap and I think it seems like there are more cases and that’s just understudied.
If there’s a topic there, then we should do something here. If I was directing the NIA and I had a billion dollars, well, I wouldn’t have a billion dollars a year because we’re trying to invest in aging.
Eric 01:00:59
Million a year.
Martin 01:01:01
Maybe I would find a sneaky way to appropriate all the gerontology money and then I’d have $500 million. How am I allocating it? There’s a thing here that at least could be very important and no one’s doing it. Let’s put some money there. Retrotransposons, chronic infectious diseases, extracellular matrix changes. These things we should do. And then also, of course, allow for people to do whatever they’re excited about, just plant some seeds over there.
Eric 01:01:36
There’s an interesting relation to our earlier point that aging may best be viewed as the loss of the originally programmed cellular and biological identity of a system. In the case of infectious diseases, let’s talk about the multiple sclerosis example. The leading theory on why multiple sclerosis seems to be linked so strongly to Epstein-Barr, probably causally determined by Epstein-Barr virus infection, is because some of the neoantigens that are presented by the virus after infection are very similar in their protein structure to myelin sheaths.
After your body’s immune system is adaptively trained against this new viral antigen, it also has cross-reactivity with your own myelin sheaths. You start to have an autoimmune reaction against your own myelin sheaths, and that’s the origin of multiple sclerosis. In that case, this loss of identity is almost like the blurring of identity between viral and self protein. There are probably countless examples of that.
1:02:23 How does loss of cellular identity relate to aging and disease?
Martin 01:02:40
If we go back to how we were talking about things before, maybe you could—I’m going out on a limb here—say that for the correct notation of biology, there needs to be a time aspect to it. The correct unit is some kind of interaction. I’m not creating Newton’s laws live on the podcast.
Daniel 01:03:05
That would be great for views.
Martin 01:03:06
Only time will tell. But the correct unit is an interaction, because what’s happening here is not that the virus is going and messing something up. The virus is changing the interaction between cells and cellular components with the rest of the body. Some trigger happens, and then forever after, you’re screwed.
It’s like if your revenue leader keeps all limos, first class flights, just wasting money or stealing money. That company will super overcorrect and everything over $50 needs approval. Now you’re just ruined. You’ve ruined something through that interaction unless you have a way of clearing it, of resetting it.
There is a way in which you could describe it as a loss of identity, but there’s also a way in which you could describe it as the creation of a particular interaction that is destructive. By putting things in a certain state from which arises this destructive interaction.
There are some theories of this. David Sinclair will say information theory of aging, which is some version of there’s information in the epigenetic stuff and then you can get it back. I am not aware that it has tried to explain intercellular aging. Michael Levin has another thing of goal-directedness. There are some things that are about this loss of information. Even the plain DNA mutations is a version of that. You’re losing information.
But why does your kidney start failing? Do any of these theories have an explanation for how everyone gets older and all the different things start going wrong, but then you get specific diseases? Uri Alon has some system that tries to explain that through the normal function. He has this systems medicine book, and it’s like, what do the cells normally do? There are different types of cells, and their normal activities, how often they divide and so forth will describe it. But we don’t have much in that realm.
Of course it’s a hard thing to form a theory around because there are so many possible interactions. Then you could start thinking, what is the experimental approach that allows us to start mapping those interactions with the correct unit of understanding? Like seeing that you could modularly start putting stuff together.
It’s not like when I did chemistry in high school. This was not my topic. Everything was just memorize a different thing. I didn’t get it.
Eric 01:05:59
That’s a shit place to be in. Protein names is a similar thing. PRK1A, ASK1. You’re just memorizing stuff. It’s not that useful. How do you get to the point where you see where this stuff flows? You’re Magnus Carlsen. It’s not just chess pieces. He probably sees the flow. I don’t know how he visualizes it, but sees what will happen from the state of what is. He has the chunks.
Daniel 01:06:27
I was thinking about if you picture the manifold of potential states for a cell or for a whole organism. It’s not just a random landscape. There is some healthy equilibrium and then there are other stable equilibriums that exist that are particular diseases.
Aging might involve all this randomness that is occurring, but there are particular stable states we’re going to end up in. The reason we have these states is because we’re an organization where, like you were saying, one department in the company is overly spending, so then a new rule comes out. They do find some equilibrium. You find a new state of organization. It just ends up being a state that, well, maybe it’s not even stable. It could then decay, or it’s a state that’s just generally dysfunctional.
Martin 01:07:16
Most of the states with aging, most of the disease states, lead to decay. Your heart will get worse and worse, and then maybe your kidney gets worse because it’s no longer getting enough clearance because the blood flow is too weak and so forth. Most of them decline, just like most companies will by default decline. If they are not growing, most of them are declining. There is some sort of, now we’re failing gradually. We’re not really winning. Then the good people leave. Now we’re winning even less and we can’t get out of it.
But as you said, it’s not just some random constellation where all the cells are just doing totally random stuff. It is the system trying to self-correct. Often when you succeed at self-correcting, you get some infectious disease and often you’re just fine afterwards. Or someone stabs you or whatever goes wrong, and you just end up mostly 98% fine afterwards.
When you are aged, that is less likely to happen. We know that the human body has built in—I mean, this is life. If you don’t have any kind of homeostatic capacity, then you will just die because the environment is changing often. Every organism has homeostatic capacity for sure. The aging part is the loss of that capacity, which could happen either because your pieces are just shitty and physically full of some kind of junk or something. Could also be because they are busy and they are actively responding to some perceived state that could be real in the case of a chronic infection, or maybe not real in the case of some sort of inflammatory loop that you’ve gotten into that isn’t useful at all.
Daniel 01:08:59
On the topic of homeostasis, the organism, a human, needs to maintain homeostasis. Each individual cell does, given perturbations around them. As you have disease progressing in other parts of your body, you’re creating more pressure on each individual cell to survive in this new environment.
Martin 01:09:18
Right.
Eric 01:09:20
A great case of that is fibrosis. You have this feedback loop that is locked in and progressively worsened through the changing of the extracellular matrix in the course of fibrosis. The fibrotic environments instigate cells to become more fibrotic in their extracellular excretions as well. A negative feedback loop that happens in the wrong direction.
Martin 01:09:41
Fibrotic diseases definitely are like that. Some other diseases are different. They’re degenerative, and maybe have a similar dynamic where as some cells die, the other cells have to pull harder. That’s just more demanding. You’re like, okay, an all-nighter sprint, whatever. Eventually you quit just along with everyone else.
1:10:07 Why are there no trillion-dollar biotechs and what would it take to create one?
Daniel 01:10:04
Let’s talk about another state of an industry: biotech. Biotech is not doing great. Lata recently wrote an article, Where Are All the Trillion Dollar Biotechs? She talks about Eroom’s Law, which is every nine years since 1950, the cost to develop a drug has doubled. We’re now at $2 billion to get a drug to market today.
She talks about some of the issues the field is facing. We have different candidates that are meant to inflect the curve around the cost to develop drugs: genetically validated targets, drug repurposing, and AI. But she argues it’s not going to get us there. We need large markets like aging and we need other things to change.
You recently wrote a response to this article. I’d love to hear your take on this problem.
Martin 01:10:54
The large market part, there’s clear empirical evidence. The closest thing we have to a trillion dollar biotech, if you’re willing to call pharma biotech, is Eli Lilly at $700 billion-ish. They’re going after a large market. Maybe we will get there that way.
Eli Lilly is also a very old company. They’ve been around for many decades. How did they get there? It wasn’t like the cool way. I don’t know who’s cool, Cursor, Tech Stuff, Deal. I don’t know what these companies do, but very rapid ascent.
Daniel 01:11:37
Radicals podcast.
Martin 01:11:40
The Free Radicals podcast. Cash challenge laid out for you guys.
Over a long period of time, they had multiple successes and now they have a very big success and they’re able to capitalize on it. There’s a large market, and that’s good. Those drugs are going to go off patent. What will be Eli Lilly’s market cap after those go off patent? You don’t want to be a trillion dollar company for like a week. You want to be a trillion dollar company and then a $10 trillion company probably one day. That’s really what we are talking about: companies that just keep growing and growing and becoming more and more influential.
Some of the dynamics of the biotech industry are directly opposing that. One of them is that when you make a drug that is good, this is Eroom’s Law. What are the explanations for it? People are like better than the Beatles. Once you make a drug and it’s pretty good, now the next drug has to be better. How hard is it to find something? Unless you’re getting way better at finding the best thing, then it gets harder and harder. The patients disappear. At the macro scale, if your market is shrinking the better you do, that’s somewhat perverse.
Eric 01:13:05
Perverse incentive.
Martin 01:13:07
If you’re doing Stripe or whatever and they’re like, okay, we’re going to make money off these transactions. The better our thing is, the easier it is for someone to start a company. They have Atlas and everything, making it easier to make a company, basically just growing their user base. The user base keeps growing, they make money the whole time, and they make more and more money as the user base grows and their product probably also becomes better. Then they can capture more of it. It’s better, better, better, win.
But in biology, oh, I’ve solved this problem. Now the problem is, hypothetically, if you totally solved it, gone. I must find a different problem to solve. How close is that new problem to your original problem? It’s curing a disease, but it’s very different biology. You’re starting from scratch. You’ve fixed diabetes. Does that tell you how to cure Alzheimer’s? Not really. You have to start over repeatedly.
You don’t get to capture value for your initial solution for a long period of time. The value that we generate in biotech is for the good of humanity. It’s humanity knowing, it’s very David Deutsch. Humanity knows how to do a thing that we could not do before. That’s the actual value.
People are like, oh, this drug costs $5 to produce and they’re charging $1,000 for it. That’s crazy. But it’s not production. It’s not cars. It’s not iPhones. It’s not like you couldn’t physically screw it together yourself. You have no idea what to do. You don’t have the concept. Maybe like an iPhone in that one, but the first one, you didn’t even know you wanted this thing. You had no concept of it.
We have done so much work, and those $2 billion, it’s not like to run the trials and so forth. It’s like doing all the science and doing all the things wrong and figuring out what do we even do here. That is the value that we generate in biotech.
The social contract is that if you do that and you tell others how to do it, then you get a patent. Your patent will give you typically on the order of a decade where you can sell this and other people cannot. That’s where generally you make money. 50% of the money from drug sales comes in the US. I think it’s like 80% or more comes in the patent-protected period.
You solve a very hard problem. It’s hard, but many things are hard. It’s hard to build AI or something. But then after you’ve solved it, you can only make money on it for a little bit. Now you have to solve a new problem that is very dissimilar. That’s one big issue.
The other big issue, probably even a bigger issue, is when do you get rewarded? When do you find out if you are correct? This is like, oh, we can do CRISPR, we can do generative AI for molecules. All of that stuff doesn’t matter that much. What matters is what happens when you put it into the human. Do you get the drug approved and make money?
If that takes a decade after you’ve done some cool new thing, you have no feedback. The cycle doesn’t work. 10 years? Imagine if Stripe built this payment system. After 10 years, the FTC or something will approve that people use our payment system.
Martin 01:16:22
We will find out if it’s really good at that time. Please continue to fund our software engineers in the meantime so that we will make money and so forth.
Eric 01:16:35
In business parlance, the cash conversion cycle is very long.
Martin 01:16:38
Very long. You’re solving a lot of different problems. I forget if I called it in the end, where are all the billion-dollar biotechs? There might be trillion-dollar biotechs if we really wanted them. That’s another way of putting it from a systemic perspective. The industry has taken all of this information and then decided we should do things in discrete stages, and we should disassemble the problem-solving team once their problem is solved and it’s now a different kind of problem.
That will take a long time to figure out. You get some scientists to do some great biology and figure out a hypothesis for how to treat this disease. Then you start doing clinical trials. There’s going to be a long time. Investors will directly or indirectly suggest that you should maybe fire all of those people and just run your clinical trial because your clinical trial could work and it probably won’t.
1:17:32 What are the key components needed for a biotech flywheel?
Eric 01:17:46
There’s really a few things that we’re talking through here. The first is the perverse incentive of treating a disease after you’ve already treated it well is increasingly harder and more expensive. The second is the idea that all early stage discovery is essentially an independent process from one program to the other. You’re almost always restarting from scratch in some capacity, so there’s just a lot of startup costs that you’re eating up as an industry.
The question is, how do we move past this? We can recognize some of the deficits of the industry, but what sorts of levers can we pull in the course of the next year, the next five years that can dramatically change the rate of discovery or the rate of commercialization or just the mechanics of the industry?
Martin 01:18:31
The rate of discovery is not that helpful, as we sort of alluded to. Obviously, it’s good, and that would mean more things get tried. One thing you could say is, well, what do we need a trillion-dollar biotech for? Just every time, break it apart, start a new company, do a new thing. It’s fine. That could be one answer. I think that’s maybe an implicit answer for some biotech investors. This works fine.
Daniel 01:19:04
Let’s argue against that. One obvious point is that the IRR for biotechs is below their cost of capital, which is something Lotta mentions in our article. Also, obviously we want flywheels. We would like the ROI instead of being down to be going up. We’d like to be getting better and better at treating disease.
Martin 01:19:29
I think that’s right. You would argue for a trillion-dollar biotech in the scenario where that company creates more quality-adjusted life years by dollar, or life expectancy per year, or these kinds of metrics that are the point of doing it. Human freedom from disease, human existence enriched.
What is the trillion-dollar company that is better at doing that? You can invert it and say you won’t get to it. That’s kind of the point of the essay. You won’t get to a trillion dollars unless you have a flywheel. So where is the flywheel in biology? When you do something well, it becomes easier for you to do or cheaper.
One thing you could say is, can you reduce the value capture delay so that you actually get feedback? Because right now you don’t get direct feedback. You get this boardroom feedback. Oh, we think these scientists did a good job. They seemed good while they were doing it. We haven’t measured yet whether they did a good job at each stage. Can you push that further backwards?
I’ll lay out three things, and then we’ll get back to how would you do each one. That’s one. The other one is, how do you make your solution fit to more ideas? If you are, what is it really that you are creating? Are you creating a way to understand diseases? Are you creating ways to create new molecules? Right now, you trend towards creating that which you get rewarded for.
Gordian is a company that finds the way to treat a disease intrinsically. But because success rates are very low, that is not very valuable until it has been validated. The bread and butter currency of biotech is a molecule, something that has patent protection and that you can put into clinical trials. Everyone’s trying to make molecules faster.
1:21:43 How can biotech companies de-risk clinical development?
Eric 01:21:45
This is why we have the kind of current landscape of biotech, which is me-too iterative assets that are additions or updates to prior molecules.
Martin 01:21:56
There’s a couple of things going on at that stage. One is there are some modalities that are more modular than small molecules. Small molecules, this is your traditional, every time you go make a new thing, you’re like, oh, I want to do exactly this, disrupt this interaction between these two proteins. It’s very start from scratch. Can we make that less start from scratch?
One company went on some trajectory and then it went back down is Moderna. They’re like, oh, we’ll put whatever into our lipid nanoparticles. If we need a new protein, we’ll just encode a new protein the way that we got the COVID vaccines very quickly.
Daniel 01:22:37
Vaccines work.
Martin 01:22:37
The way we got COVID vaccines quickly was the technology was compatible with that. Now we just put a different thing in. Once you discover what works, now you can do a thing very quickly. Alnylam is a bit similar, but with siRNAs. As we shift towards modalities that work potentially better than small molecules, which is a steady march of progress, that’s enabling those things. You could do more of that. However, you need a way to find out what to do effectively.
Eric 01:23:04
On that point, there’s this concept I’ve been spinning up for a while that someday I’ll formalize a little bit more, but this idea of platform escape velocity. Platform biotechs are allowed to become their own independent entities the likes of Genentech or Amgen or Alnylam when they’ve reached some point of accretive value generation from additional asset creation, where each additional asset is decreasing the marginal cost of assets.
In the case of Alnylam, for example, they did something very critical, which is they pioneered this ability to specifically and efficiently target the liver with siRNAs through GalNAc conjugations. Once they deconvoluted that particular problem of actually getting the genetic medicine to the liver, at that point they were able to unlock a number of different genetically validated diseases that could be treated in some capacity with siRNAs. That’s now their bread and butter, hitting the liver with these.
Martin 01:24:02
siRNAs.
Eric 01:24:02
Right. In the case of Moderna, I think the real challenge is that they had a bit of a black swan event with the mRNA lipid nanoparticle technology for vaccine presentation being really good for a problem that happened to pop up, which was COVID-19. But after that point, there was a bit of a search process of what else can we do with this particular technology, the constraints and the possibilities of it. They pivoted a little bit more towards cancer vaccines, which I think is still interesting, but as we know from oncology approval rates, a much harder space to find traction in and much more variable.
I don’t know where my whole point with this was, but this idea of platform escape velocity is incredibly hard to pin down for any space. I think that is ultimately what you’re looking for. How can you find some sort of space where you found an edge that you can continue to build improvements in?
Martin 01:24:53
But you need to span the whole value creation, value capture process. Let’s say that Alnylam finds a way to hit the liver well with GalNAc, and then they just start curing liver diseases. There’s X number of genetic diseases, and they just go through them, and they will actually become very valuable. Right now, they’re a little bit valuable, but not very valuable. How do they then keep doing that?
They need to either, which is sort of what Lilly is doing, make a bunch of money and then just start buying up a lot of targets, or they need to invent a liver platform. They would benefit greatly if they were also Gordian and had created the liver disease atlas and you had your manifold of how hepatocytes behave and you can put many different diseases. Now you have like, I find new things to do with high efficiency, and efficiency that increases. Gordian, as we map out more stuff, we have more of the manifold and it gets easier to do it. There’s some amount of flywheel, but then you have to be able to turn that into drugs over and over and over again, which may mean not small molecules and going into one of these modalities that are still maturing. But that may be another component.
Then you still end up at that clinical trial rate. I think as long as we are just like, it’s a gamble, it’s a gamble every time, and as long as that’s the default state of investors and most people involved, I think it gets hard. If you’re profitable, maybe you’re fine, because now you can just make your own bet and say, well, we’re just going to keep doing all these trials and we’re pretty sure they’re going to keep being right, because we believe in this early stuff. If you are actually right about that, now you’re capturing the whole thing. There’s a long lag, but you can shorten it. It’s fine, because you’re just going to cure everything.
Fourth component, this is why it’s hard, you need a market that stays or grows. I think this is something that was at least insinuated in Lara’s essay. She says you should go after age-related diseases because those are the large markets. I rephrase this slightly and say like, if you’re going after not treating a disease that has manifested, but you’re going after making the human body more optimized, either in the sense of you could go to just enhancement stuff, or just in the sense of, as things are continuously going wrong, you’re figuring those things out and just fixing them proactively, constantly. This has its own challenges because preventative medicine payment, how do you get reimbursed, all this kind of stuff. But in principle, if you keep solving something where the complexity of the body exists, it creates new problems and you’re getting paid to solve them constantly. You have all those things. Now you can do it.
I think you can substitute, duct tape in some of the areas. There are still many diseases that are not cured. If you can just do it repeatedly across diseases, but you need each of those four steps where you have a way to fund the trials, you have a way to figure out what to do, you can turn it into something that you’re doing, you can fund trials, and you can keep doing that over and over again. You need that whole loop to work without a reset where it’s like, new disease, start over, none of these siRNAs work anymore, GalNAc is not useful anymore. The more you capture the whole thing, I would say you definitely become a trillion-dollar company. If you can do two or three, maybe that’s good enough. I think we’ll find out with Lilly and Alnylam and others.
Daniel 01:28:40
Thinking about Gordian, Gordian might have all the makings of a flywheel. But the way you judge a platform is through clinical approval. Let’s say the platform is so good and it has all these flywheel, it has the makings of the flywheel and maybe has a 99% hit rate on the therapies it develops. But you go to the market and you’re unlucky in your clinical trials. It’s the one out of 100 that doesn’t work. Is that the type of thing that then blows up your flywheel because you got unlucky at the start? How do you think about—
Martin 01:29:12
I think that happens often. There’s two ways around that. One of them is common, and one of them is Gordian can do it, maybe a few other people, most people maybe can’t do it.
The common one is that you just hand off the risk. Go back to Genentech. What did they do? We have a platform. Our platform can produce protein drugs better than everybody. We are going to partner with Eli Lilly or whoever to do this thing, and then we get income from that. We will just go further and further along the value generation chain and get less money than if we had made the whole bet ourselves. But we can do it repeatedly because the platform is sort of modular. We will keep doing that until we get to the point where we are profitable ourselves. This has happened a number of times. All of those first wave of founder-led biotechs that you described, like Genentech and Regeneron and so forth, did a version of this, where there was a bunch of money coming in from partnerships. That’s something you can do. I think that’s something that Gordian will do. We create more assets than we can clinically fund. We will probably partner at different stages.
The other way that you can get away from this, roll the dice, is to run clinical trials as not a single shot on goal, but a cluster of shots on goal. You could do what’s called a basket trial. Basically you can put four drugs into one trial. Each trial you’re going to have your standard trial, you’re going to have a control group and then you’re going to have some number of doses of your one drug. Often you’ll have one control group and then two doses. So now you have three arms. If you add another drug, you’re adding two more arms. You’re not quite doubling the cost of the trial. If you actually had two drugs that had different reasons for potential failure, now you could de-risk—
Martin 01:31:18
You could create a portfolio within the thing. I think that’s something that Gordian potentially can do because we are in the business of finding new targets. Many companies are based around, this is the target we have picked, we will find a way to do it. But then you have your target risk still correlated. If there is no target risk at all because of genetic disease, maybe that’s a good idea. Maybe you’re like, we’re going to do a gene therapy and a protein enzyme replacement thing and whatever.
I think there are some companies, Unicure I think does that, or Genzyme. No, what’s the one? BioMarin does that. Maybe BridgeBio to some degree. I don’t know how much internal competition they have, but that’s another way that you could do it. But of course it needs a bigger chunk of money up front.
I think there are a lot of the dynamics that happen in the biotech ecosystem that hypothetically could be more efficient if we were optimizing for the system. But unfortunately, there are individuals involved and comparators involved. Some people can go raise $800 million or something. Most people cannot. Most investors have some amount of money that they are able to raise for their fund, and then they want to take that money and deploy it across 25 companies or whatever. That means that each company for their trial gets this much money, not 3x this much money.
And then also the people involved, everyone has a lifetime. Everyone’s making decisions at the end of the day that are like, let’s say you had a 1 in 10,000 chance of making the best cancer drug ever or making Ozempic. Do you want to take that chance? Maybe, but you only get 3 chances in a non-assisted lifespan. Would you rather take 3 chances, expected value for each one is whatever, 10 years of healthy life, but you fail in all but 1 in a million cases? Or would you rather take a, I have a 10% chance success, and if I win, all I get is to be millionaire and everyone thinks I’m cool. You have to be really idealistic to go for the other one.
Now, if everyone went for the other one, we’d be great off. But that’s another challenging part of it. You need more crazy people. I think most people doing biotech are crazy a little in some ways. It’s because, as you said, the expected IR is low. You will mostly fail a lot and then not get rich. But if nobody does it, then we all just have no options for— every time something goes wrong, there’s nothing you can do.
Think of the experience right now when you get sick or someone you know gets sick, you go to the doctor and you hope that there is something that can be done. You got diagnosed with something, depending on how much you know, if you don’t have a PhD in biology, like, is this a treatable thing? I have pancreatic cancer or I have prostate cancer. One of those, you’re probably fine. And one of them, you’re probably dead soon. The existence of a human being to be like, there is something we can do. We can take care of this. Or at least that’s what I want for me. That’s what I want for you, for everyone. There’s something we can do. If none of us do this, then there’s nothing we can do ever. It’s just leeches or bloodletting or whatever, eat some random tree and see if you don’t die.
Eric 01:34:53
The very controversial take that I have is that, there’s this general sense that medicine should be effectively a public service, it should be free, that these sorts of things should always be just given people, no questions asked. I have the opposite view. These things should be even more expensive and the companies making them should get even more wealthy and the employees making them should get even wealthier because the number one problem of biotech is not that the science doesn’t work, it’s that we simply don’t have enough money in the system to incentivize the best people.
Martin 01:35:22
Right. Biomedical R&D across the entire world is like $400 billion per year, I think. Or let’s say, just take US. So then it’s like $250 billion-ish. And healthcare spending per year is like $5 trillion-ish. So 20x. That seems not ideal.
How do you fix that? There’s a lot of things now that are complicated. Basically, the primary market for all medicines is the US. And that’s enabled because the rest of the US healthcare system is really expensive. The US drugs are more expensive because it’s not that big a deal because everything else in the US healthcare system is so expensive. The drug costs are only 8%. I think they just tolerate it because it’s still not that big a deal. But how do you get at that whole— I don’t know, that’s a hard problem. I don’t want to solve that one.
1:37:02 What is Norn Group and what problems does it address?
Daniel 01:36:22
I’m grateful that you are crazy enough to work on biotech and working on some of the hardest problems. And you’re not only running Gordian, but you also have a nonprofit, Norn. Can you tell us about that?
Martin 01:36:33
I think as a person, I’m just unable to exist in a state where I’m ignoring something that’s broken. That’s why I started working on biology and aging. It’s also probably an advantage of being a founder, or at least would be a strong disadvantage if I was anything but a founder, because I’d see something broken somewhere.
Gordian is doing something that I think is important and increasingly important as we get more me-toos and more clustering around a few targets, which is find the way that we scale, industrialize the way that we find out what to do. But there are other problems in the whole ecosystem, and we’ve talked about a number of them. At some point, I was just like, we need to solve everything. I don’t see everything being solved. I will have to do more.
I started spending my Sundays on this nonprofit that I would say the essence is that there should be a plan that feasibly allows us to win in longevity. I play a lot of board games. Generally, if I don’t have a plan, then I will probably not win. If you’re playing sports, I did football when I was younger. If you come in and you just have no plan and people are just doing stuff, you will probably not win. If there’s a discrete win that we want, let’s say that within our lifetime we can actually make the way that you feel be equivalent to like half of how many years you’ve lived or something like that. Let’s say that we want there to be an answer to every disease, whatever version of that, then we need some strategy that will actually get us there.
Norn Group is a nonprofit do tank that tries to first lay out what is the system, what are the drivers of progress and the constraints, and then what could we do that would change some of those. We talked, now we’re going full circle to the beginning of, well, this appears to be the system. What are the levers that we could pull in the system? What are the things that make a really big difference?
Throughout our conversation, we’ve arrived at because most clinical trials will fail, there’s only so much funding. That’s even worse for aging. No clinical trial has succeeded. $250 billion for biomedical R&D per year in the US or $400 billion globally. Aging field globally is like $1 billion. It’s less than 1% of biomedical R&D. That’s very low for something that is the primary risk factor for many of those diseases, most of them.
Why is it low? Because nobody thinks it’s going to work, or very few people think it’s going to work, and we’ve never shown it to work. We’ve just bullshitted for like 5,000 years or whatever. Things are different now because now we can reliably make animals actually live longer. We’ve done some things that are actual truth-seeking. How do we unlock that faith? And how do we do that without having to do dozens of clinical trials for aging drugs?
That’s happening. Unity tries, that doesn’t work. New Limit will try and Retro will try. Gordian will try. We will keep trying. What else can we do? That’s the kind of thing that Norn does, is come up with where is the highly leveraged, something could be done that would strongly improve the odds of success in the field. Look at what is being done. Do we need to make sure that people know about epigenetic reprogramming or partial reprogramming? No, getting lots of attention. Some people are doing—one way you could prove success faster is do it in pets instead of humans. Loyal and other companies and nonprofits are trying to do that. Those are all being done. What’s not done?
It would be nice if we could measure aging without waiting for aging to happen. It would be nice if we had epigenetic age test or a clock or whatever. It doesn’t have to be genetic. It could be proteomic. It would be nice if we had an aging clock. Well, apparently we do because there’s so many papers on clocks. Are any of the clocks predictive? Like, could we just substitute the clock for a lifespan experiment? It happens sometimes in papers, and I don’t know about you, but for me, I’m just like, whatever, ignore. I don’t trust it enough that I would do that.
At the societal scale, clearly we don’t trust these epigenetic age tests enough to make big—like, if an insurance company—insurance companies care a lot about whether people will die. If they were industry-grade, whatever. If they’re good enough that the average sophisticated person believed that when you get like this, your biological age is 24 or whatever, that’s true, then insurance companies would mandate that these go out everywhere. We would start running clinical trials. We are not there.
Why are we not there? Well, one thing that has never happened in the 13 years since we started publishing on clocks is to do a bunch of different treatments, prospective test of the accuracy, specificity, and sensitivity of the clocks to longevity interventions. Just take 20 groups of mice from the intervention testing program where we have a dozen things that extend lifespan and many that don’t. Treat the mice with different things, blind test. Can any of the aging clocks tell me without knowing what treatment I did, is it—how long is this mouse going to live? Why have we not done that experiment? How much does that experiment cost? Maybe like between $1 and $2 million if you want to do it really at scale and spend some money to put up a public accessible data infrastructure. Anyone can submit a new clock, the data is free for everyone. That is a thing that could be done for a small amount of money. That should be done.
That’s my soapbox at the moment, but it’s just sort of like create the big picture map, make sure that there is a plan for the field that could result in success, and then just continue taking the next move that maximizes the probability of success according to that plan. Gordian is an important thing. There’s many other people who are doing important things like Loyal that would substitute there. And then some things are yet to be done, like creating benchmark clocks. Or funding other understudied areas like we talked a bit about earlier.
We should have—there should be a—we did a program called Impetus Grants, which was this fast grants for aging, try to fund high-impact research by having competent people. But instead of review by committee, it’s sort of like max not average score, do it all in like 3 weeks so that scientists don’t have to spend all their time on bureaucracy.
Martin 01:43:47
145 grants. It’s resulted in a bunch of papers, some clinical trials, and so forth. That’s a program that’s happened. We’ve had three rounds so far, thanks to a bunch of philanthropic donors. Juan Bonet was the anchor for the first one. Robert Rosenkranz and James Fickle and Vitalik Buterin have put money into that. It’s a 1% overhead, very efficient way of just like, hey, I want to support aging and I don’t know exactly what to do. Then we deploy it to stuff that potentially could be really changing.
I want to do future rounds on chronic infections and aging. The people who are bravely going and doing that research, put some more funding there. That’s going to give you an outsized return. That’s a good thing. The benefit of the aging field being really small is that there’s actually quite a lot of opportunity for individuals, whether financially or through their talent, to make a meaningful difference.
If there’s on the order of 2,000 researchers that are doing stuff, and some are better than others, if you’re really good and you’re in the top 10% of those, you joining the aging field could be like half a percent or a percent more progress. Just one person. If you have $1 million and you’re like, we should make sure that biological age tests work or something, you could do that thing. Compared to some of the amounts that are happening in various industries, you actually have an opportunity to do a lot.
1:43:58 What opportunities exist for individuals to impact the aging field?
Daniel 01:45:22
Where can people learn about you?
Martin 01:45:25
Norn, we have a website, norn.group, N-O-R-N dot group, and there’s a Substack. It’s called Life Expansion. You can also find it via Norn, where there’s a lot of the writing. And then Gordian.bio is the company website. And then I’m on Twitter. Martin B. B. Jensen is the handle. It’s Borch, not Borscht. I am not Russian soup. Maybe after aging progresses, but the rest is not Russian soup. A de-identified soup. That’s what we’re hoping to avoid. Please help me help you avoid all of us becoming Russian soup.
Daniel 01:46:08
I’m on board with that. Thank you.
