🔬 "The Most Innovative Diffusion Research Is Happening in Drug Discovery, Not Image Generation"

I remember very clearly in like 2017, 2018, talking about GANs and how generative adversarial networks and how, uh, they're clearly the future of image generation. Obviously, they didn't work very well for for proteins or protein ligand systems, and we sort of had to wait for the right primitive to get created, and that turned out to be diffusion, which turned out to be a much more useful primitive for the space. What's kind of cool is right now, for people that are are interested in really core fundamental AI research, actually some of the most innovative diffusion research is happening in our field, is happening in 3D structure prediction right now. No one would have predicted that then, but now, like, that's a pillar of diffusion, I'd say.

Hi there. Welcome to the Latent Space AI for Science podcast. I'm R.J. Haniki, CTO of Mirmix. I'm joined by my co-host, Brandon Anderson, and we're privileged to have Evan Fineberg, a founder and CEO of Genesis Molecular AI, and Sergey Udov, who led Llama 2 and Llama 3 pre-training before he joined Genesis as CTO.

Hi, I'm Sergey. I studied physics at school. This is a long time ago. Um, and after graduation, I happened to work in software engineering. Thought I will never need physics again. When ML came up and became a thing, and turns out that a lot of things you're doing in ML were actually very similar to what you would do in physics. So I jumped in on the ML band, and I did a lot of AI research being a part of the FAIR team in in Facebook, uh, for quite a while. I later on led the Llama team, um, Llama 2 and Llama 3 models, and then recently I decided to pivot my career again and, um, recover my roots in physics a little bit, and I joined Genesis as my CTO.

Hey, I'm Evan. I'm the founder and CEO of Genesis Molecular AI, and, uh, like Sergey, also also a physics major. I was a bit different from everyone in my family growing up. Almost everyone was a medical professional of some kind, and, uh, my sister became an accomplished TV writer, playwright, novelist, and my love was physics and computer science. Uh, but my mental model of an adult was, well, you should help people and help patients ideally. So, um, I was always searching for the right, the right way to do that, and, um, after arriving at Stanford, doing my PhD in Vijay Pande's lab, we were cited in the mid-2010s of everything amazing going on in machine learning, to use a dated term, uh, for for images and for language. Um, and around the same time that that Sergey was at FAIR working on a lot of big graphs, I was at Stanford working on many small graphs. Molecules are really networks of atoms and bonds and spatial interactions. And if you're at the right place in the right time to bring to bear our backgrounds in in physics to improve AI algorithms for looking at molecules, uh, we published a few papers in the area of graph machine learning. And sort of like Sergey, I thought, well, I won't need this physics again because, you know, there's machine learning. But turns out the massive amounts of simulations on GPUs of proteins that we ran also came in quite quite handy as Genesis evolved, and we've been really excited the past few years to figure out how to build foundation models for a totally new domain and make them useful for patients.

Yeah, that's a really nice lead into my first question, which is, I, you, we've both been in this domain of sort of machine learning for molecules and bio for roughly 10 years. It's kind of an entire generation of tech bio has, you know, come and gone since then.

While a lot of machine learning for molecules has been, I think, quite effective, one domain where it's been not effective has, or historically really resisted, uh, machine learning modeling has been the world of protein small molecule interactions. And with some of the recent advances that Genesis has put out, it seems like you might have actually started to make real improvement on this in a way that we haven't seen for a long time. Can you talk about what you have done, what Genesis has done, the developments which led to improvement, and why you think this is actually a real improvement over kind of some of the traditional, uh, machine learning strategies which were ambiguously helpful?

I totally agree, Brandon. It's, and the amazing thing is, or one of the, one of the really remarkable things is that when we were founding the company initially as a spin-out of the research we were doing in AI at Stanford, um, we were afraid that we could be too late. I mean, you remember that you and I first met around that time, right? Seven-plus years ago, as you rightly point out, almost a different generation in the area. And at the time when we were a little seed company, there were already incumbents in the space who had raised orders of magnitude more money than us. There were already some really exciting academic papers that were out, and we had to constantly answer the question, sort of like, is there room for another company doing AI in drug discovery? Which is a crazy thing to say, uh, looking with, with a backward-facing light. And, you know, hindsight's 2020, I suppose. And the claim that we made then is the same that we made, we're making now, which is that there's 20,000 protein-coding genes, and each of them can can cause a disease. And in the same way that in other domains of AI, it's not like they were zero to one and then suddenly that problem was solved. The reality has been there's been no single iPhone moment. Um, even the phone required iterative development over time to become what is today. And people love to talk retrospectively about the ChatGPT moment, but the first ChatGPT models that became widely used were vastly less useful than than they are now. And I think we should look at drug discovery from an AI context in the same way that, as we go, we should expect to solve more and more problems. And there will be large epithets, there will be iterative improvements, but they'll compound over time. And that's why I think in the same way that 10 years ago was very early, but clearly value was starting to be created for drug development then from an, from an AI perspective, fast forward 10 years, our technologies, which we'll talk about in this discussion, are vastly more useful than they were then. In the next 10 years, we'll see another exponential improvement as well. In the same way that your car, in terms of autonomy, can do things today. We're, we're recording this in San Francisco. You look outside and and see see Waymos, we'd be blown away, you know, a few years ago by by those technologies. So I think we should both be looking at it as every few months, every year, the technology gets more and more useful for discovering new medicines with artificial intelligence, and that will continue to to be true for years to come.

Maybe we can go back in time a little bit and explain, like, what was the state of protein small molecule drug discovery, let's say about a decade ago? Like, what could we actually expect to do with machine learning models, and what were some of the failure modes that people, I think, didn't really see coming from a practical standpoint back then?

It's funny, just, I mean, breaking the fourth wall a little bit. I, I, I came in not, not really knowing any of these questions, but Brandon really is knowing how to speak my love language because, because like, like the other people in this room, you know, though my my background is much more in the the physics and CS side, I've been very, very focused in a deeply passionate way about the area of how can we make medicines with AI for over a decade. Though I've seen so, so many tectonic shifts. As a view, to give you one concrete example, and, and I'll give a little background for those who don't have as much biological background. Drug discovery is akin to finding a key for a lock, where the lock is usually a protein, some cases a nucleic acid, right? Um, and where the drug is the key, and it's usually some some small molecule or a peptide or or or an antibody or some other modality that wants to bind to that lock, that protein, the receptor, and change something about it. If it's an enzyme, you want to usually stop it from functioning. Sometimes you want to activate that endogenous protein, but it's to introduce an external molecule, an exogenous molecule to change something about that biological pathway. A necessary but not sufficient part of that process is finding a molecule that binds well to that receptor or that protein. Of course, that's not sufficient. We also need to make sure it does not bind to certain anti-targets, certain proteins you want to avoid. That often mediates toxicity. You need ADMET, which is typically like 30-ish properties to make sure your molecule is safe, in addition to being effective and gets to the right tissue. Uh, these are all critical and all are necessary and none are sufficient. But one concrete example to answer your question about protein-ligand interactions is that we had a hypothesis for a long time that if we can predict the 3D structure, the 3D coordinates, can imagine for the CS people, like a point cloud, um, the 3D positions of your drug and the 3D positions of the protein. And if you could model that with high accuracy, that will lead naturally to measuring binding affinity or or predicting potency more accurately. That was a hypothesis that fundamentally could not be tested because models for predicting those 3D poses, the 3D structures of complexes were so bad.

Or if you can predict this, it requires so much computational tech resources that you might as well just solve the problem.

Right, like you might as well solve the 3D crystal structure or cryo-EM structure, is what you're saying, which is can cost tens of thousands of dollars, um, it can be months or years, right? Entire PhD thesis, postdocs can sometimes work literally 24/7 trying to solve one 3D structure. And the hypothesis was that if we make this orders of magnitude faster, sometimes more accurate, which we can talk about with artificial intelligence, or before that with with molecular dynamics, that you would thereby not only accelerate drug discovery but enable the discovery of medicines for targets that were previously thought to be undruggable. And that was just a hypothesis. And it, and it took until the last few years to show that if we systematically improve those the accuracy of those those predictions, if we can prove the accuracy of a protein-ligand 3D complex, we can make potency prediction more accurate as well. And that's something that I am personally extremely gratified to feel like we've been showing because that was just a hypothesis when we started a company a few years ago, and, and now it's become a reality thanks to the convergence of a lot of different ideas that were able to to to enable Pearl and other foundation models like it.

Yeah. Can you talk about Pearl a bit? Um, so the, I think one of the interesting things about Pearl in my mind is it was an attempt to kind of scale what seems unscalable to me. Uh, RNA, small molecules, you know, normally if you scale them, models just come out as just like pure pattern matchers. These things love to tell you what you already know, and sometimes they're maybe not so great at telling you something you don't know. So maybe it seems like there's hints that Pearl has actually moved beyond this. Um, like, what were the key insights which went into this?

Sure. And by the way, Brandon, like what you describe, that shouldn't surprise any, like, pure ML practitioners in the audience, just because machine learning is initially constructed for pattern recognition. And it's the same with when you fire up Claude or ChatGPT, or what have you, it's going to do best when it's closest to the training data, and it's going to pattern match off it. So I think that has been the big sense of urgency in AI meeting the physical world is how to extrapolate, how to make generalizable models, and that's, we've been really hard at work on. Maybe, maybe Sergey wants to give some more more color on that, maybe to step back a little bit and discuss what Pearl is. It's a structure prediction model, which basically means it takes as an input a protein sequence and a ligand representation that you try to attach to this protein, and it predicts how this protein and ligand are going to look together, uh, as a structure in 3D space.

So, like a term for this people sometimes I've used is like co-folding. It's multiple folding. AlphaFold 3, and then some of the, and then Bolt and, um, OpenFold 3, and so on, have also implemented some of these ideas in their own way.

Yes, there are, there are several models, some are open source, some are closed source, that are pursuing similar, similar directions. Our fundamental difference is that we're focusing on the small molecular space. A lot of models that are doing protein-protein interactions, and it works well for small molecular space. On the first glance, it may sound like it's an easier task because, like, well, it's a small molecule, right? So, like, why would it be challenging? But the reality is the search space is so vast. All for small molecules, there are 10 to the 60 drug-like small molecules in the universe. Uh, so good luck searching for the space. And there are also all of all variations, like you can rotate it. You can have different conformations of those molecules.

So, maybe the intuition there is that you have, like, when you type in a Google search, if you just do a few words, you're going to get a huge list of matches. But if you have a very specific query, then you're going to be, you know, you're going to have a few matches, and they're more likely to match. So, similarly, with a small molecule, if you have a very small query, which is your molecule, then it's likely that you're going to have to search a huge space of of possible matches. Whereas if you have something very complex, then there's, it's easy to rule out a a match at that. Is that a good way to think about it?

That's one way to think about it, definitely. Uh, for me, it's just the the computational complexity on figuring out which small molecule will attach to this protein.

It's such a vast, um, problem to solve that is impossible to do without good models.

Mhm.

It's like finding a needle in a haystack.

Yeah.

Where everything except your needle is very, very dangerous, or or or just doesn't bind.

Or they are needles.

Yeah. Right. Right. We're finding hay in a needle stack might be a more apt analogy.

Yeah.

Yeah. So, so now, now we can go and think about how we train those models, right? So, a lot of training data that people are used is so-called PDB, uh, which is a public database of like all of historical crystal structures, and it's not that big. It's like 200,000 crystal structures, and it's very hard to expand. It takes a lot of, uh, time and energy and money, uh, to create new crystal structures. Although there are some projects that are pursuing this, it's still expanding at a glacial pace. Um, so expanding this database is pretty hard. But what we figured out is possible to do, and it's very relevant to our small molecular space, is that in small molecular space, you can actually model your small molecules with physics. Uh, you can model their behavior, and that allows you to create more data. But you can train a model on something which is not necessarily possible in protein-to-protein, uh,

Because they're too complex for a protein.

Yeah, those are very big molecules. So it's very hard to model them with physics. Not impossible. It's just computationally very, very hard.

Right. So, so I guess what you're saying is you do a really good job of modeling these small molecules using MD or something like that, and so you can kind of put together those structures in your training set at a much lower cost, and so that you can use that as your synthetic training set.

Maybe, maybe to step back a little bit about our and talk about our roadmap. Um, so it's not fundamentally different from the roadmap for LLM scaling. Like, remember in LLMs, we have all of the stages. You have pre-training scaling, then you have post-training scaling, where you do like either fine-tuning or RL. Now everybody's doing RL, and then you do inference time scaling. So all of those three concepts, they connect, and that, that's what led us to state-of-the-art LLMs these days.

It's not fundamentally different on our side. Uh, we also have pre-training scaling, where we create a lot of synthetic data to train better models. Um, we do that. Um, then the second thing we we started doing is actually the third step in LLM scaling. We started doing inference time scaling. And it's fundamentally very similar. Like in LLMs, when we talk about inference time scaling, we're talking about thinking tokens, where an LLM, instead of giving you the answer right away, it kind of goes and thinks for a while, and then comes up with a response, right? Well, we are doing very similar thing with our models, where a model is forced to think, except it's not thinking in language tokens, it's thinking in terms of crystal structures, not fully materialized crystal structures, but some sort of a crystal structure representation in memory. Um, and the model kind of goes back and forth with those. Um, and we use physics-based guidance during this process to steer the model output in the right direction. And what we found is that it improves model performance by a lot.

It's easy to understand how thinking tokens work because it's just basically parts of the transcript that you don't see. But I think that, at least for other models, do you have like sort of some sort of loop in? Is that similar to what you're doing, like a loop transformer, or something like that, or is there, I know you mentioned that there's some physics-based verification as part of that in the loop. Is there more to it than that?

One fundamental block of our models is diffusion-based head. So it's like the same diffusion models what people are using for image and video generation. Well, we use them for crystal structure generation, right? And diffusion head is by nature iterative. It's like multiple steps. You're refining your predicted structure. Um, and as you're doing this process, you can steer the model in the right direction.

So there's like, you have like a steering simulation or something that in the in the loop in that? So you, you go through a passive diffusion, you look at what comes out, you run some verifier and use that to say, I like this or I don't like this, or give a vector to go towards. Is that kind of the idea?

Yeah.

So, so it's like your diffusion head is basically learning something like a force field, and that you're kind of balancing out between a diffusion-based force field and like a physics-based force field, or is that a way of thinking about it? Or

I struggle to understand like what those models are really learning underneath, and I think the whole idea of explainability is a big research challenge. Yeah, you can identify what a neuron in a transformer potentially thinking about, but it's really, really difficult problem. So we, we have some internal representation, but at the output comes out as a crystal structure, but it's really hard to tell what exactly is happening inside. Is it possible to do something to the effect of the standard mechanistic interpretability things like sparse autoencoders and the more complicated things in this context, or is there a problem with doing that kind of thing?

I think it would be interesting in the research as a research direction. That's not something we are pursuing. I guess potentially we could explore this.

Okay. Yeah.

Sergey is a lot more credible than I am talking about the LLM translatability to our space. Sergey led the Llama 2 research team at at at Meta when when when he was still there. So, he's trained one of the most widely used language models on the planet. But I guess what I can add on the on more the physical and interpretability side is I like to look at at AI models in terms of the inputs, the model themselves, and the output. And, um, we find it really improves outcomes as well as interpretability to your question when you could include as many physical priors as possible without, of course, you know, biasing the model to what we puny humans believe. But I've always had the view that AI fundamentally is representation learning. And to me, that's actually not that different than what's happened in the language and the vision domains, where we're enforcing a real human prior on images. When we started using convolutional neural nets, is we're saying, well, pictures are grids of pixels. There's something inherent about that. That's that's a human construct, and we constructed CNNs on top of that. And language as sequences of tokens in one dimension. We first built RNNs, then built transformers, but it's still, that's baking in a pretty, pretty serious prior on how to to look at that data. And we don't view it very differently here. In our case, from the input perspective, the disadvantage that we have in our field over the more traditional domains of AI is that we don't have the internet to work with. Like we can't just download Reddit posts and, you know, buy some subscriptions to the Wall Street Journal and like train a model, and voila, the pre-trained model works fairly well. The closest equivalent that Sergey mentioned is the RCSB Protein Data Bank, which has more like a couple hundred thousand crystal structures or structures and others. Although there's a lot more, there's a lot of tokens per structure, so there's a lot more latent information that that figure would describe. So we have to be clever on the input side, which is generating more pre-training data, as was RJ's mentioning, um, using physics as much as possible. In terms of the model architecture itself, the idea is to not have the model have to relearn as little physics as possible, so is less likely to overfit. And then at the output stage, um, enforcing enforcing physicality. I think that also goes to one of our main, uh, focuses as a company, which is we've been focused from the beginning on small and medium-sized molecule discovery. What does that mean? So small molecules tend to be, uh, drugs that you can take as a pill. So it's orally bioavailable. It's how most people think of medicine. And medium-sized molecules. So think about macrocycles, peptides, um, modalities that break the traditional Rule of Five but are still growing, uh, growing modalities in medicine. That said, of course, small molecules are still 65% of FDA-approved drugs. So we're talking about the biggest part of the pie. We're building this with a judicious focus from day one on what do drug hunters, what do medicinal chemists, what do CAD scientists, what do they need to to discover drugs faster and better than they could before? And so we wanted to to build that sort of, um, human not only interpretability but usability from the beginning, and also physical usability. We want outputs from these models to be used by physical methods that require 3D coordinates to actually make sense. Use the term force field. We want the outputs of our model to be useful as inputs to a force field. There was this, a few papers that came out, one was in Cell, I think last year, which showed that for all the claims about AlphaFold solving drug discovery, people tried to take AlphaFold-produced protein structures, use them for traditional docking, and found no value in it. Those pockets just weren't high-resolution enough. They couldn't be used by physical screening methods. And so we wanted to be able to build our systems so they'd be useful by humans, useful by all the many adjacent and powerful tools from the physics and computational chemistry communities from day one, for to have that interoperability. Sense.

Yeah. So this might be a good time to just step back for a minute. When you develop a drug, there are many steps to that, from discovery and toxicity and availability. And so there's names for all these things, which I'll let you, you state, and then there's the clinical stuff, and finally, uh, eventually approval, hopefully. Where do, do your models sit in that, or which parts of that process do they say? And what, what are the steps, just like briefly? I've read that there's 12 steps, maybe I don't, 12 steps in drug discovery, acceptance is the last one.

Except that your phase three trial failed.

Exactly. Yeah. So where do, where do you, do your models help people to do their jobs in the discovery or in the development process?

As you rightly point out, if we go from from the very beginning to the very end. First, we need to identify what target is causing the disease, what signaling cascade, what's going wrong, what's causing the phenotype of the patient. And that's not a trivial exercise, but, um, there are many, certainly not all, but many diseases where the cause is actually very clear. The most clear are the the monogenetic diseases, but even in others, it's from various panels, it's clear that this overexpressed or overactive or deleted gene or protein is causing the disease. So we know many, but not all, of the universe of causes of diseases. Then, once a target is identified, we have the process of finding a drug for that target that's ideally selective, very potent. Um, it gets to the right tissue. Some diseases are multi-system. Some are very specific to to a certain organ or tissue. You want to make sure that drug can get to that tissue. And then there is the process of GLP tox and IND enablement. IND, Investigational New Drug. You have,

GLP in this case is not good, not not,

We're not good lab practices.

Thank you. We're not talking about the metabolic space right now. Um, there's there's clinical trials, usually ascending size, phase one starts typically more more with safety, um, then phase two and phase three, and, and as you point out, ideally approval at whatever regulatory body, the FDA or EMA, or or what have you. Um, our contention is that the highest leveraged application of artificial intelligence is the drug discovery and drug design process. And the reasons are that even though they're, they're all valuable, and I cheer on all the peers, many of whom I know who are working at all the different parts of the stack, but I think the first thing to make clear is just like a a vision model is going to be very different than a coding model. They might share some similarities under the hood, but requires real focus to do any one area correctly. Um, there's a similar, if not greater, difference from the sort of models you need for target identification, i.e., biology, to drug discovery, to preparing for regulatory filings, to helping segment patients for clinical trials. These are all very important, complementary, but ultimately distinct problems. Um, I think about all of the times where a patient goes to a doctor, gets told that they have a very clear diagnosis. And in addition, the physician will tell them, there's been breakthroughs where we understand why this disease happens. We've sequenced your genome, even. We, we know, or sequenced your tumor. We know it's causing your condition, but we do not have a selective therapy. There is no precision medicine for your condition. And it has the hope that, well, at least they know what's wrong with with me, but they don't really know how to treat it. And our aim is to, um, have as many moments as possible where patients are told they have a very specific condition, and there's a specific treatment for them. That's going to require bending the cost curve of discovering developing new medicine, but also importantly, solving those cases where there's certain quote undruggable targets, undruggable proteins that we need to figure out how to drug, where those targets have have proven resistant to traditional methods or even intractable in some cases.

Just to clarify my understanding, there's the discovery part, and then there's the clinical part where, and so there's there can be effort there where, okay, I want to find what's the target that is impacting this disease the most, or how do I maybe, what's the thing that I want to go after in order to solve a medical problem? And then there's the clinic where you're answering, does this thing actually work? And maybe there's a feedback loop between them even. But if you don't have the thing in the middle that's like, "Okay, here's how we actually build this drug, it so that I can't hit my target, it's not selective enough so it like kills cells or does things to cells that I don't want it to do." So then it doesn't matter, right? Like I can have the answer, this drug should go after this target, but then if I don't know how to do that effectively, potently, etc., then then the answer doesn't matter. Is that kind of what you're saying?

Yeah. R.J., if you, because people love to debate about success rates in in clinical trials, but the reality is, if one focuses on those drug candidates that are aimed at proteins that have a close genetic linkage to a disease and or at least where the biology is well understood, where the animal models translate well to the disease, where the molecule is predicted to have good pharmacokinetics, as and the levels in the blood in patients are in serum are going to be high enough. Those molecules have fairly high FDA approval rates. The success rate from phase 1 to end of phase 3 is substantially higher than average. People love to cite the 10% success rate, but that's really a lowball. Um, because we often know what genes, what proteins, what targets are causing the disease. They're just really hard to drug, or the molecules we put into in patients are inferior to the the the target product profile that that they deserve. In the cases where, if you just look at the preclinical data, you focus on good targets with good biology that are predicted to be distributed well and have good safety profiles, those molecules are very likely to get approved and it create tremendous value for patients. And so our view is, I think about the physicians that run trials that are clamoring for those kinds of molecules. I think about the patients that are looking for more selective therapies. And so our view is that is the highest leverage application of AI in healthcare, medicine more broadly.

Just a business question here. What is the space of things which are that the target is the biology is understood, the structure is understood, and that you know the only thing you really needed to do is find the right lock? That seems like you probably would have already picked over that class of targets in some sense. Those are easy targets, right? So how, how much is opportunity is there for that?

So there's two orthogonal concepts. Known biology is orthogonal to the ease with which one can drug that target. Sometimes, unfortunately, it seems they anti-correlate in that often it seems that the most appealing targets from a, a validation perspective seem to be really hard to drug.

No, but do they anti-correlate, or is it just that we have picked all the low-hanging fruit which do solve both those? I, I think it is likely the case that many of the quote easier targets, for a variety of reasons, have been sort of many of them have been drugged, not all. But even in those cases, um, people love to to sort of, um, have this this false dichotomy of easy versus hard targets, and well, this one's drugged versus it's picked over, therefore it's not. We, at at Genesis, do most of our work through large pharma. Like, most of what we do is providing AI services, basically, to to to major pharma companies like, like Gilead. And we recently announced our, our expansion of of our collaboration with Iktis, which we're excited about. And I can't give details of specific targets that they're working on, but what I can say is there, there's a real range from first-in-class chemical matter. What that means is, we believe this target causes a disease. There's no known molecules that bind to that target. We need to find the first binders ever, the true zero to one cases. But there's a, um, a wide variety of targets we work on that are more, I'd say, one to 10 cases, where sometimes there is an approved agent, or sometimes there, um, uh, there's molecules that are only preclinical, but they're not optimal, where if you can improve upon those preclinical molecules and get them to development candidates, they're ready to get into patients, or if you can improve upon the existing clinical or approved agents, you can create a lot of value for patients. I'll give a public example. We're not working on this target at all, but you just look at the the progression of ALK inhibitors, ALK inhibitors, and you look at, to make it very concrete, charts of patient survival. You might have said, well, when the first ALK inhibitor came out, why do we need another? We've drugged ALK. We're done. But it is a qualitative, not just a quantitative difference. When you when you compare patient survival curves from first-generation ALK inhibitors to later-generation ALK inhibitors, of how much longer those patients live, how many more get benefit from it. And so I think there's not only value in first-in-class, there's enormous value for patients in best-in-class too. And, and I think both are areas that that we focus on, if that makes sense.

Okay. Yeah. So we, so far, have talked about Genesis modeling in terms of, let's say, this one specific problem of protein-ligand binding. How does your overall world work in terms of developing this? I mean, I assume that you must, I mean, I do know that you have other things that you've worked on in solving the broader early phase drug discovery, um, problem. I keep joking about that when the Nobel Prize was given for AlphaFold 3, a lot of people thought that like drug discovery is solved.

And it's very far from the true, um, very far.

Yeah, you can, maybe, maybe so, maybe you can predict crystal structure, right?

At low resolution.

At low resolution. Well, we're, we're doing better, I think.

Yes.

Um, assume you can predict crystal structure. Does it mean drug discovery is solved? No, obviously not.

Yeah.

The broader community, like protein folding community and drug discovery community, immediately said there's other things you care about, like dynamics, like just a single static structure is not enough to understand what's going on, interactions. Um, yeah, so there's lots of things there in addition to just like having a static structure. I think many people thought those were very useful, like they've really accelerated science, but it was very clearly a starting point and not an end condition.

Right. But if you look outside of like people working in the field, like general popular community, it was a pretty wild wide belief that the problem is solved.

Now, in reality, besides,

I just have to clarify that, otherwise, like, there's an entire ML structure biology community will just jump on us and say like, "No, no, it's not a solved problem."

Got to, got to throw those caveats in.

Absolutely not a solved problem.

Yeah.

In reality, you need to predict so many other properties, uh, like ADMET properties that I haven't mentioned. Like, basically, you're designing a key for a lock, like that small molecule that sticks to your protein in the body. But you don't want this small molecule to stick to everything in your body, cuz that's probably going to cause issues. You also want to make sure that small molecule can be soluble, uh, so that you can ingest it as a pill. Uh, those are important properties too. You want to make sure it doesn't have any other side effects. Um, and predicting all of those properties is also just as important as, or maybe even more important than, predicting the crystal structure itself. Um, and of course, as a company, and we're very proud of it, um, we're not just building models for predicting crystal structure. We're building models for predicting all of those properties and basically enabling drug hunters to kind of become 100x more effective in in their daily job.

I'm curious about this, the pipeline that you have. You mentioned you have different pharma partners, GSK or whomever you're working with. Do they have targets that were sort of drugs that were designed by you that are in, like, clinical trials, approved? Like, where, where do we stand with all that?

We're limited in what we can say about pharma companies. They are, are famously secretive, and which makes sense. What I can say, which was recently publicly disclosed as an example, um, is, um, we just expanded our partnership with Iktis, and that started approximately, you know, a little over a year ago, the initial work together, and that spanned the two sort of bookends of the drug discovery process in some way. So, one of the, the, the programs we worked on together was a case where a very challenging target, where chemical matter, um, existed, but there's a key binary event, which is getting to a DC or development candidate. And that is the molecule, or ideally a set of molecules, all of which are possible to be, um, uh, the agent for a phase one clinical trial. And, um, we had to to do some work together where we, we take our foundational models, our base models, we fine-tune them on Iktis' data and use them in, in a close collaboration to to sort of work together to to get to a DC faster. And we're getting, um, substantially closer in, in that case, which is, um, one of the reasons we're excited to expand our work together. And on the, the flip side of that is one of the other areas that we worked together was on a protein with also very nice linkage to a very severe disease where there was no known chemical matter, like no patents, no papers, in this case, no co-crystal structure of a ligand, another synonym for a small molecule that bound to that protein. So we had to find the first ever known chemical matter to bind to it, and then we progress those, what are called hits. Hits are like the first molecules that that bind your to your protein. Progress those into, um, inhibitors that are active in called biochemical assays, which are more enzyme-based, and cellular assays, which means that in an actual cell, living cell model, um, your molecule is active as well. So it was really based on, on, on a concrete set of accomplishments together that we want to expand the collaboration. And unfortunately, outside of that, there's really limited any versions of what we can say. But, um, the objective of Genesis is to create medicines that patients wish they had. And the way we'll be able to do that is by working with as many pharma and biotech partners as possible, for whom their comparative advantage is discovery, clinical development, commercialization, and our comparative advantage, of course, is in AI, where where we can put those, those two expertises together in in a synergistic, not just an additive way, and, and make medicines together that otherwise would would not be possible. Yeah. Um, so that's the name of the game that we're in this for, is those those clinical outcomes that you're pointing out, and I'm very excited to be able to share those, um, as, as they rise in the coming years.

So one thing that is, you guys talk about on your website and whatever is this one angstrom threshold at which a protein structure prediction and the binding between that and a small molecule becomes useful. Can you talk a little bit about, like, okay, why is that the case? What, how did you do it when others have failed? And, you know, and you've spoken a little bit about that with the, you know, sort of biases in the model, the synthetic data, maybe there's more. How does that impact, you know, sort of downstream, the actual things that we're talking about here with, you know, uh, the ADMET and all that stuff?

When we're talking about molecular interaction, the scale of two angstroms that people typically measured is just too big. So you can think about, like, in image generation models, you would generate a picture, and it's like fuzzy.

So, yeah, it's like, in general, right? But you can't discern the details, and the details really matter here. Like, with two angstrom accuracy, your entire aromatic ring can be flipped, and it will still be, uh, a valid, uh, output. The worst part is that unlike an image which is blurry, uh, you, you don't even know it's blurry, right? You flip, you flip around an aromatic ring, and it looks just fine. And,

Yeah, or even worse, a heterocyclic aromatic ring, and then you're really in trouble.

Yeah.

And it really matters in this case because, like, individual atoms need to establish connections here. Um, so that level of accuracy, what we're pushing for, uh, one angstrom, is really important. So maybe it's almost like with a lot of generative image models, they will fill in a detail which just doesn't exist. So if you were to try to do, like, forensics, and you tell, you know, your, your vision transformer to to enhance an image, and suddenly it just pops up a face, but that's actually not the person's face, right? But, like, you don't know that, and it feels just fine. And so if you're trying to use this as a structural hypothesis, as a med chemist, you're kind of screwed. If you're looking at the wrong face, as an investigator, it's, it's not going to take.

Yeah. Yeah. I really like your framing. And then, as Evan said, at, like, as a downstream, you run a bunch of other models, like physics-based models, and all of those little issues, they compound. Uh, and then obviously, if you made the wrong prediction, like fundamentally wrong, then the downstream predictions are also going to be wrong. I just wanted to say for your question, I really appreciate your mid-2000s TV reference about, you know, enhancing forensic images. So, I just want to say I appreciate that. Before we get back into back into details here,

We talked about poses. So, maybe the contrarian take is that a pose isn't even a well-defined concept, and that, you know, the best way of thinking about these things is this like, you know, probability distribution over things. And there's a small molecule which probably lives in a binding pocket, something like this. I mean, is this pose, like, this is an abstraction that humans use and not really, in some sense, ground truth? So I'm actually almost kind of interested to actually hear that, like, oh, we have this warning threshold. This is like really important for med chemists. I would also talk to some chemists who would say, I mean, this pose isn't even real, versus just like maybe a most probable configuration? Or, you know, how do you think about that abstraction in general, and do you explore, do you explore conformational space and provide that as tools? And, yeah, how, how do you interact with just like a single structure versus like an ensemble? And sorry, and I don't know if that question even makes sense then.

Yes, it's an abstraction, but it's a very useful abstraction. It helps us to build up confidence that a particular,

Model output is actually valid.

Mhm.

Uh, but you did not just straight up hallucinate something.

Cuz yes, ultimately what matters is binding affinity or potency, and you can straight up predict that with your model and just skip the entire pose generation step.

But then you only have a single number, and that number might as well be completely hallucinated, and you have no means to validate whether that number even makes any sense. So, as much as poses are not perfect,

They're still a very useful tool, uh, for the entire process.

Sorry, going into that correction. Um, you know, just, I'm afraid I'm jumping into a technical rabbit hole here, but, um, but like, just because you have a pose also, I mean, there's like entropic and enthalpic contributions, and predicting binding affinity is not just about getting the energy right, but actually, is this even likely to make it into the binding pocket, and is it likely to live there long term? So it's much more than just, you know, affinity, potency prediction is much more than just, is this the right pose, but does this have the broader properties it needs to be a long-lived molecule in this state? And how do you kind of deal with those things too?

A note for sort of the, the wider AI audience that probably resonate, both both with practitioners as well as users, is, you know, obviously something that's blossomed in the in the past six months is agents.

And we love agents. Who doesn't? There's a lot of, I'm going to say, a few necessary but not sufficient conditions. I'd say, in my, my sort of response, if you're really asking a deep question there. We all remember what agents were like, let's say, mid-last year. And let's just say there is positive value and there's negative value. And both can be amplified by agents. And why is that? Well, agents are only as useful as the underlying models that they're orchestrating. And let's think about coding. If your coding model even makes subtle but real bugs, your agents are just going to amplify those issues, and you're going to end up with not only slop but something that may be anti-useful, that might give the user incorrect information. We all remember what that age was like mid-last year. That made a lot of people lose patience, I would say, with claims about LLMs for agent engineering.

Something changed. Clearly, a threshold was met. And even though these models are still not perfect, the utility of LLMs for software engineering is so obvious now. I mean, obviously, it's been a huge tailwind and very useful for us for replacing a lot of the drudgery of coding and getting it focused a little bit more on some of the more strategic issues that matter. You could draw a direct analogy from that to what we're talking about here, in that we obviously, it will be no surprise, but we're working on an agentic platform for 24/7 drug discovery. You can imagine just fleets of hundreds of med chemists and CAD scientists working nights and weekends all the time for your drug targets. The codename for that gem is Sapphire. The prerequisite for that was we needed the underlying models for pose, 3D complex prediction, potency, ADME to all be good enough for an agent using these models 24/7 to create molecules that medicinal chemists would actually want to make and not laugh at. And if your model is sitting at 1.8, 1.9 RMSD, that's slop.

Most likely, and let me be really direct about it, what you're hearing is the biodal distribution from people of the utility of a 3D pose and how real it is. The reality is that for a highly potent ligand, almost certainly there is a large portion of that molecule with a very well-defined 3D position down to even half an angstrom. And if you don't believe me, you can open up the electron density diagram in the PDB. It's all available online, and you can literally see in some cases aromatic rings with missing density in the middle. So it's not just a construct on a blackboard that your organic chemistry professor showed you. It is, you can literally see a donut, a torus of electron density for an aromatic ring. However, there's going to be solvent-exposed areas, often times, that will be less important for binding affinity but maybe are important for solubility or other properties of your molecule. Those sometimes are less defined because sometimes, in reality, to your point, they're more dynamic. They're flopping around in solvent. But the critical piece as an upstream indicator that's valuable for predicting the free energy of binding is to get the core of your molecule that specifically interacts with the protein to be correct to sub-angstrom resolution.

And why does this matter, just to use it intuitively? So, a hydrogen bond, everyone in the audience has probably heard about a hydrogen bond. It's the most critical form of non-covalent interactions. It's how nucleic acids are held together. It's how most ligands and proteins interact. It's how proteins form secondary structures. Hydrogen bonds have a very specific angle and distance. And the distance is from the donor to the acceptor heavy atom. It's 2.7 angstroms to 3.3 angstroms. And if I do my math right, that's a 0.6 angstrom gap. And outside of that, it's not a hydrogen bond. If it's less than that, it's a clash. If it's more than that, the interaction is much, much weaker pretty quickly. And an angstrom, for those who don't know, is one-tenth of a nanometer. So drug discovery really is a science of resolution. And if your accuracy is not sufficient, it will therefore not be useful for the downstream things that you care about, which is both potency prediction but also prospective design. What molecule do I make next? So that's why our view is that the history of innovation in startups is that the ones that do best are ones that focus on one well-defined but very important problem. And we think our ability to get higher resolution predictions stems from our judicious focus on small, medium-sized molecule design rather than boiling the ocean.

Yeah. So that's the what and the why. So what about the how? How did you get to one angstrom and sub-one angstrom?

I'm going to give you an extremely boring answer.

Okay. Uh, which I think is actually true for the entire AI field. Like, where three things matter in AI. It's data, infrastructure, and evolves.

Right. So you can only improve what you measure. And once you are very careful about measuring what matters, and you have really talented people on the team, we're going to figure out how to hill-climb that measure. Right? So from the start, we actually focused on less than one angstrom precision, and that led to a bunch of small decisions in the process that compound. Right? If your team never looks into this metric at all, then you will never train a model that's good at it. And if you're constantly looking into that, then you're going to achieve that.

So it's the right objective plus good science. And it propagates everywhere, right? Through the whole stack. It propagates to how you look at the data, maybe like how do we filter out data. Some data is more noisy. So maybe we don't need to see it, or maybe you don't need to see it later in the training. It propagates through your modeling architecture and propagates through your loss.

How long have you been working towards this specific goal? I'm curious. This is not something that you broadly hear in the community talk about wanting. People do say to have decided one angstrom is sort of the canonical benchmark, and I've never heard someone say one before is the cutoff until like reading things coming out of Genesis. How long have you decided this is the number we need to hill-climb on? How direct has this focus been in the evolution of the company? I'm just wondering like, how did this come about?

I think one of our most important secrets is that we are working on real drug programs, either with partners or in-house. And when you actually work on real drug programs, you see the failure modes, you see what works and what doesn't work. And it's pretty obvious when you see at the outputs of real programs, what kind of failure modes are happening. And it becomes very obvious that one angstrom is just not working out for those setups.

Right? But I mean, there's a lot of really smart med chemists who are also, you know, who think very carefully about benchmarks and people I respect very much and who also have prosecuted actual drug discovery programs. So I'm, I'm just kind of curious, like, why hasn't this become part of the community? Is this literally just that the community has never been able to succeed at a winning benchmark that they kind of settled it too, as, okay, it's something we can aspire to? Or I, and I don't know, just...

But I mean, to your point, it sounds like there's, in my experience with pharma, there are plenty of problems for which people, there's kind of these known things amongst the technical experts in a subdomain, but that it doesn't get out of pharma, or it doesn't get the attention it deserves because, you know, for whatever reason, some of the information is proprietary and gets kind of passed from company to company but never really released into the public. Is this your estimation of what's going on?

Have you ever heard of SWEBench? So, um, Gemini does pretty well in SWEBench. Sometimes Gemini publishes models that win on some of those software benchmarks. Raise your hand if you're using Gemini to write code right now instead of, you know, the obvious other name competitors.

Yeah. Not me.

No one. Like, why would you do that? It's obviously worse in practice. Um, and I'd say actually that in this case, if you look at the provenance here of how that happened, RMSD less than two came originally from, again, B and I've been in the space forever, from docking studies, long before AI models for pose prediction. When physics-based docking studies from academic institutions, because usually large, the proprietary software makers, they didn't want to benchmark their methods against other methods. So academics had to get licenses and try it. And then they'd introduce "less than two," which is not surprising. They're academics. They're not using these things to make drugs. They're using it to write papers. And so that got, sort of, the provenance is that that kind of got repurposed by the AI community. But the actual trend is much more in the direction that we're saying. So the first, uh, big innovation there was PoseBusters, we'd say, put out by a lab at University of Oxford, which pointed out that RMSD itself is insufficient and we need to look at physical validity as well. So that's PoseBusters, I'm talking about not the benchmark, but the metric. So Oxford improved that. And in the latest release from OpenBind, we just published our benchmarks at Genesis on OpenBind set. We'll talk about that in a bit. We plan on, but if you look in their, you know, that's their original publication, their default metrics are using RMSD, PoseBusters, validity, and LDDT. So there's clearly an acknowledgement that's already been made that RMSD less than two is insufficient, and the field is now rapidly evolving to acknowledge that.

You think that the academic literature is going to establish some benchmark? Maybe it's one angstrom, maybe it's some LDDT or some other metric that will kind of converge and then hopefully, as a community, drive forward these more accurate modeling strategies?

I would say there was an eval crisis in our field that is now in transition right now. As our field, which was previously, let's say, a lot more quiet, but if you go to NeurIPS and ICML every year, year after year, the workshops get a lot more crowded. Those evals are in transition now that we realize the flaws of what came before it. That was a really interesting point about, you know, pick the right evals and then just do good machine learning like you normally would do. But I kind of picked up on something Evan said earlier, which is that you started out with, I guess, what we call potential net. I don't think we defined that earlier, but you know, this graph-based network, and then you did a lot of computational simulations kind of after that, and now you're sort of going back into, now once generative modeling took off. I'm curious about how that evolution worked, and how did you find computational techniques to be really crucial to sort of building on the ML for a while? And did that sort of computation lead back into generative, or did you start collecting data to build a generative computational data? Is this a way of data documentation for generative models? Like, what's sort of the history of that, and then what led to those decisions to the extent you can talk about?

I will say that the last line of PyTorch I've written is much further back in history than the most recent line of PyTorch that Sergey has committed. So I want to make sure that he gives his opinion as well. Um, I want to make sure that I'm going to directly answer your question. I want to make sure that I address something you said a little bit of time ago that I think kind of got lost, which is you asked about how we do other things, including ADME prediction. So, I just want to make a comment about that. Just, I can spend all day and I'm happy to diving in details about Pearl and the evolution, and we will. Um, I just kind of want to give it a shout out to the fact that it is one important pillar, but not the only pillar that matters in the trajectory of a drug discovery campaign. And you mentioned that potential net paper, which we're happy to see had become influential. Since you and I were working in the space when it was very much in the future, let's say, but now we, now it is the future, so it's great. Another paper we published around the same time was on neural networks for ADME prediction. And we published two, actually. ADMET, can you...

Oh, sorry. Thank you. Yeah. Absorption, distribution, metabolism, elimination, and toxicity are the five prongs that comprise ADMET.

These are all the properties you talked about before that you just have to get right in order to make a drug successful.

Correct. So I would say there's over 30 or so assays, each of which you can imagine, if you're a neural net person, like a multitask neural network or multi-head, it's got to predict over 30, you know, three dozenish properties, each of which if it's in the wrong range means your molecule is not a drug, it's just a tool. And so these are things like solubility, which Sergey mentioned, important for formulation, as in, can it be made into a pill that you can take orally? Your oral bioavailability, whether or not you're inhibiting certain enzymes called the cytochrome P450s and its different variants. hERG, a critical channel which if you inhibit it too much can cause cardiotoxicity. So these are an alphabet soup of things that most people here haven't heard of.

And a lot of these are extremely hard, specifically because it's not like a single causal effect. It's, there are often times many processes, many pathways which are involved in defining this one endpoint. So...

Correct.

And data sets are often times comically small.

Unless you have, you know, I guess pharma to help you out, but at least open source it's...

Hard problem.

It's sparse in the public domain. I'd say there's actually a range from really directly predictable properties to ones, let's say, that as you point out, are actually amalgams of other signaling events. So like whether or not you're inhibiting like CYP3A4, that's really specific. It's a certain protein that you're inhibiting.

So that's something that Pearl, for example, could actually model and expect to see some performance here. Yeah. Useful prediction.

We've been doing 3D work on ADME prediction before anyone else was. And there was a bunch of work that ended up launching launching the company. Back in the day when AI could still be in peer-reviewed journals, not just sort of like random white papers on Archive, we published MoleculeNet, which has been cited a few thousand times at some point. And we also published this paper showing that multitask graph neural networks were the best at the time for doing ADME prediction on large pharma datasets. And that paper is one of the most cited papers on AI for ADME ever. Now, a lot of papers, as you know, it's like you write them and you kind of move on. But that work, both those works have become quite influential as well. So our history from the beginning has been to work on not just one problem, but to focus on drug discovery. And in so doing, being able to have the bandwidth to focus on building all of the ML models that are needed for drug discovery and not get distracted by the biological discovery processes that are needed, the target ID side, or the clinical trial side, but really focused on all those tasks that are required for drug discovery, and also molecular generation, right? These are all important. And I just wanted to make the point at the beginning, but as you rightly point out, Brandon, Pearl is a 3D structure prediction model, and many of these ADMET properties can be posed as those. So it is therefore useful, and not just on-target potency prediction as well. All of them matter, and we've had to tackle all of them.

Do you use Pearl for all these tests, or they're focused on some of them?

We'll be sharing some things publicly, um, in the coming period. But I think we've had enough news of results lately, I think. I think so far, um, if that makes sense. Uh, every time you publish something, it's actually, it's a lot of work for the team to put together. Uh, so we'll, we'll do that in the coming period. But the most recent thing we published was obviously the OpenBind results, which was a for standard 3D prediction task.

So, so one thing I want to ask about, I kind of keep hitting at this. So, you know, you talked about there's a lot of synthetic data, and then there's, you know, there's ML modeling, which is very kind of generative, modern, generate model flavor. I want to talk about kind of the feedback between how your computation, your ML, and how wet lab data went into developing this model. And, you know, we talked about priors. I think maybe one, it's been typically hard to scale protein-ligand models well, in a way which meaningfully generalizes. And there, you know, this recent Pearl paper and some of your other recent results, we'll talk about shortly, have shown true generalization. And what I'm kind of curious about is how do these different aspects feed together to give you this power? And specifically, you know, just saying computation is interesting, but like, you have to be very careful about computation. MD has all sorts of biases. If you're not right, it can give you poor results. And so I'm curious about how all of this worked together, and especially in the history of the development of the company, how did you kind of get to this point?

There's this classic concept of the narrative fallacy, where in retrospect, it's everything seems obvious, and you can draw really linear processes of how we got from point A to point Z. But the reality is always more interesting and much, much messier and much more non-linear. I think, and for Sergey and I, who've been training or taming neural nets for over a decade, we remember a lot of different eras in the space. And we would have been so excited if we could tell ourselves the capabilities of our technologies 10 years ago. And if we told ourselves how we got there, it might have seemed somewhat obvious, but still, there were some things that would have been really difficult to foresee. So, one example is, you know, the concept of using generative AI for the molecular space is not a new concept, but the reduction to practice would have been very difficult a decade ago. And so, as one concrete example of that, I used to co-run the Stanford AI Salon. We had fun organizing, you know, it's Stanford in the Gates Computer Science building, and we'd get some interesting speakers, and like 25 people having wine and cheese, and they're all now like, very, like, just straight-up famous people, right? But back then, AI was a much smaller field. And I remember very clearly in like 2017, 2018, talking about GANs and how generative adversarial networks, and how they're clearly the future of image generation, obviously. And so there was a lot of work then also in, well, can we just apply GANs to produce confirmations of proteins, or can we try to use them to produce protein-ligand poses? And for all the same reasons that those models were really tricky to train for images, mode collapse was the most famous problem, they didn't work very well for proteins or protein-ligand systems. And we sort of had to wait for the right primitive to get created. And that turned out to be diffusion, which turned out to be a much more useful primitive for the space. And interestingly, is actually a lot of image and video models. Some are using diffusion, but some of those have actually gone autoregressive. And so what's kind of cool is right now, for people that are interested in really core fundamental AI research, actually some of the most innovative diffusion research is happening in our field, is happening in 3D structure prediction right now. Would no one have predicted that then? But now, like that's that's kind of a pillar of diffusion, I'd say, that was unpredictable 10 years ago. Everything I just said. And in parallel to that, long before with the advent of 3D diffusion models for generative tasks in chemistry, we were building a variety of tools for the problems of drug discovery at hand. Some of which were using physics-based methods for predicting potency or even for helping predict certain ADME properties. And same with molecular generation, that is like, you know, using different techniques to generate new molecular ideas. As Sergey pointed out, there's 10 to the 60 drug-like molecules. Searching that space efficiently is hard. So we've been working on that problem. And those things happened to be available when we wanted to take what was then the very nascent area of co-folding, which was clearly exciting but not at all ready for prime time, not all ready for what a medicinal chemist would want to use in their day-to-day work, and take it to the realm of useful and in some cases irreplaceable, like clearly superior to a non-cofolding method. And it kind of just happened that we had been building those other primitives, and so they were available to us, so we could put them together to build out, for example, the synthetic data pipeline, or to use some of the inference time techniques. Like, we were ready to do those rapidly because we had been working on orthogonal techniques and ways of approaching problems. So there was something here that was contemplative, but like almost any discovery, like the discovery of penicillin. I'm not saying what we're doing is on the same order of magnitude or benefit to humanity as penicillin, but there's always this element of if you're laser-focused on a problem for long enough and spent enough time in the lab, or in our case on the computer, banging your head against the wall on these problems, you can create the luck, as it were, that will enable some of the developments.

How does the lab interact with the development process?

We, as I mentioned before, we train other models, not just structure prediction models. And for those specific models, lab outputs are extremely useful right away, like potency prediction. Those outputs can be used directly for model training. But what I'm most excited about going forward is reinforcement learning. I think that's coming in our field, and we have seen early signs of it working already with our models. So where you basically put initially, maybe physics-based feedback to improve your models through RL, typical RL loop, but eventually you can go all the way down to the lab in the loop setups where you're not, your model is not only producing predictions, but you synthesize based on those predictions, and then you measure downstream properties, and then you feed back those predictions into that model.

So how do you get enough volume? I mean, do you have automation to do that, or do you like large campaigns that aren't necessarily automated but generate a lot of data?

One thing I'm super proud of is our partnership with Insight. They're such an amazing company. They are extremely, extremely good in producing data. Basically taking the compounds, generating them, creating them, and then measuring the downstream properties, and then sending those results back. This is such a partnership born in heaven for us between Genesis and Insight, where we are able to train models, give predictions, and have results back from Insight super quickly.

So your like sort of rollouts include a lab like a lab iteration?

Essentially. Yeah. That's amazing.

Yeah. And diffusion is slow, so it's about the same speed of that sometimes. Give it a week.

It is true. It's like, I'm a big believer that companies are typically really good at one or two things and do best when they can focus on it. And in the same way, we've been just really doggedly focused on developing the best AI models for drug discovery. Like Insight has that level of maniacal focus on optimizing drug discovery and development. And another sort of trend people like to be talking about is sort of, of course, the rise of China in biotech. It's the elephant in the room that there's no point avoiding it. It's so in the zeitgeist right now. And where so many western companies have, you know, had become to rely more on CROs to do a lot of their wet lab work. You know, Insight became like so state-of-the-art in terms of the experimental capabilities they had in-house. Their productivity is extremely high. And as Saki said, it's a match made in heaven for us because what we thrive on is continuous learning of the models. So we want to have design, make, test, analyze cycles that are as rapid as possible and continuously fine-tune, in some cases, retrain the models based on what we see in the lab. And so that partnership is one of, if not the first ever, that enables that sort of true joint foundation model training on historical and also prospective data. And as ML practitioners, we all know here, the data is just such a critical input in the whole critical ingredient. So it's just extremely exciting for us. And that's a range, you know, from reflecting our conversation so far here, from structure, potency, and a variety of ADMET properties as well. So, so I think it immediately will improve the strength of our models and be really powerful for for for generally accelerating drug discovery.

And and just we were joking around about the time that it takes to do diffusion, but like, I don't know, you, I don't know if you can disclose this, but how long does it take to, you email them with here's some, I don't know, how many, 10, 100 compounds, and then they get back to you with measurements of those realized compounds in what, a day, a week, a month, an hour?

It depends. Like, some compounds are really easy to knock out, really known, like a very well-characterized reaction where the usual conditions just work. And sometimes it's, oh, this coupling actually didn't work like the literature said it would, and we have to try different conditions. So it depends. I'm saying this so the audience understands that for all the claims out there about just like robotic labs automating synthesis, the reality is just a lot more complicated than that.

Sorry. What are some of the complications there? You radically...

Radical. So they all do some form of automation. What, like, and again, not trying to get you to get into some fight with them, but like, what, what was the contrary take here on, okay, what goes wrong with automated labs?

Okay. So, this is also one of the reasons, as Sergey mentioned, we're really excited about reinforcement learning, because it circumvents the problems of having to do very fast design, make, test, analyze cycles from the lab to feed back into the model. Ideally, we can throw more and more GPUs at the problem and have the model self-train in the same way that we once did for board games in the late 2010s, or most recently for coding. Like, that's a key northstar for us. However, before anyone tells you that there will be a country of geniuses in a data room just solving drug discovery, there are some problems in our space that don't lend themselves to RL as naturally, and that will require experimental intervention. And so to drive in the specifics of why, to give some examples, and I, so I, you know, coding was always my northstar, what I wanted to do with my life. But I spent a fair bit of time myself in the wet lab as a very mediocre experimental scientist before I found my true joy back in, you know, when I got to graduate school, and I could just focus on coding all day. The reality is quite messy. So in order to make a new molecule, you have to synthesize it. And what that means is different chemical reagents and catalysts and temperatures and solvents coming together in the right way, in the right protocol, to try to make your new molecule. And once you do that, you need to purify it. If you don't do that, your positive or negative assay read could be an artifact. You need to characterize it. So you need to use NMR and other techniques like mass spec to indicate that what you made in your vial is what you actually sought to make in the first place, which is not as trivial as it sounds either. That's not true only for us, by the way, for anyone doing material science for AI. It's the same sort of challenge. Where did I make what I actually made is actually not a trivial question at all. It's easier in small molecular drug discovery than it is in materials, but it's still non-trivial and takes time. We've seen so many cases where screenings of large molecules sounds great in paper because you can test millions, billions of compounds in one go, so you have more shots on goal. And as a bonus, you have data for your model, right? You just got millions of data points. While, okay, the reality is that the translation of high-throughput screens, whether it's DNA-coded libraries or more traditional screens, the R-squared of those predictions to like the actual business of re-synthesizing a molecule de novo and doing a low-throughput, high-fidelity experiment is a shockingly low R-squared. That false positive rate is enormous for a variety of reasons. And so that's it's one of the many reasons why people are so excited about AI making those predictions, because in theory, they can be even cleaner than a lot of the wet lab work, especially the high-throughput work, and yield to higher enrichment, higher true positive rates.

So why can it be more? Is it just because the experiments go wrong a lot? And even, and even does it matter that you have some sort of very precise robot doing it? There's still too many variables. There's too much slop in the system.

It's a few things. So one is that drug discovery frequently involves finding the outliers.

Yeah.

One of the many aspects that's so wild about our space is that the properties that one is seeking in a molecule very often anti-correlate. Binding tends to improve the more hydrophobic, the more greasy your compound is, because protein binding pockets are greasy. Guess what? That makes your solubility worse.

Yeah.

And then you want to improve your solubility, often by adding polarity to your compound. And suddenly, for those that remember high school biology, cells are surrounded by lipid bilayers. Your molecule might not get through the cell membrane because you made your molecule too polar. So the properties you want often anti-correlate. And that's where you get where multiparameter optimization of molecules ends up feeling like playing whack-a-mole. So you're often searching for real outliers. And that often requires making molecules that are fundamentally novel. And the kinds of chemistries that today can be automated are fairly constrained. And so your ability to search chemical space broadly for those really top, top pareto optimal compounds is actually very limited. So the benefits you get in speed have a very harsh trade-off with the actual quality and novelty of the molecules that you can make.

So you can do a lot of boring stuff very quickly?

Uh, if you're lucky. Yeah. That would be, yeah, if you're lucky.

But to get to the, to actually the answer to your, the sort of cutting-edge questions, you really, it's hard to get a robotic system to do those kinds of things.

Yes. We would love for that to happen. It'd be a huge tailwind for what we do, but that's not, that's not available today.

Not where we're at today.

Right. And so, and so I, I mean, I know again, not asking you to disclose anything that you can't, but what is Insight's approach that makes them so fast and effective?

I think if one is willing to focus on one problem and say no to distractions, it's amazing what can happen. And I think one is that their talent density is very, very high, and they're willing to, what is not universally done now in large or medium-sized pharma, which is build so many experimental capabilities in-house.

And I actually think that it's possible more companies to try to do that, given the global dynamic of pharmaceutical development that's changed so dramatically.

Right. So CROs taking over everything, then now suddenly has become commoditized, and then that, but your also your capability to do the cutting-edge experiments goes down. It can, it's been a debate for hundreds of years about the benefits of vertical integration. It comes in all the areas of risk and challenges and upfront cost, but has all the benefits of controlling the process end to end.

When it works, you can do amazing things when you do things end to end.

Yeah.

Um, of course, it comes with greater challenges and I think an appetite, um, uh, needed to do it. And this kind of brings me to another question I have, again, more industry. But so you guys, so maybe now is a good time to talk about your pivot to from like, so there's a name change from Genesis Therapeutics to Genesis Molecular AI. And that, I think, goes hand in hand with maybe a pivot in the strategy of the company, or is that in terms of, okay, we're going to focus on the AI portion that and then sell that to pharma, rather than be a, you know, sort of a drug company with that also sells the tool? Is that, is that sort of accurate?

When we started the company, um, the origin of the company was fundamental deep learning research methods that we had developed at Stanford.

Right?

And the objective we were trying to solve for was, how can we have the greatest impact for patients? And at the time, you know, the founding of the company was purely an AI research. There was no real precedent at that time for a pure model company really making it in the realm of just creating value for partners in pharma or really most legacy industries. And the other, I think, component was the company being founded on AI research. I think we really wanted to show people that we were serious about the biotech domain. So, you know, I think that's part of the origin of, let's call it Genesis Therapeutics.

Yeah.

We seem more serious to the, to the, you know, all the tried-and-true folks in pharma, the die-hard med chemists, drug developers, who we, who we really wanted to work with. So I think there was an element of an attempt at nominative determinism there, a bit. Um, but also an acknowledgement that I don't think anyone had really shown at that time that you could be a pure AI model company.

Yeah.

And really make it in the space. And that was obviously, this is 2019. So we had the benefit of being real forebears, pioneers in the space. We had the downside of being pretty early.

Yeah. Thought to be late, but turns out it was still early innings. As the company grew and evolved, um, we really were structured kind of like a double helix, in that we had extremely strong, focused AI research. And we were fortunate to be able to recruit some of the most experienced and accomplished drug hunters, like people on our management team and board, who have discovered, co-invented, developed many FDA-approved drugs. You talk to most med chemists, most med chemists feel very fortunate if they're able to work on one molecule that gets FDA approval in their whole career. And we're really fortunate to convince these sort of accomplished drug hunters to come join our management team and board at the beginning and steer the show. So the outcome has always been the same. Everything we're talking about, it only matters if we end up helping patients and their families at the end of the day. And we have felt from the beginning that the way to do that at scale is to primarily partner with large, medium-sized pharma companies and biotech companies. So they can do what they're best at, which is novel biology, clinical development. We can do what we're best at, which is AI, and bring those two together. In addition to that, as Sergey has said, there's numerous advantages to having in-house wet lab data generation, clearly dogfooding the models, getting really direct human feedback from real med chemists, what we need to build, in an extremely candid way. And of course, the third thing is that the single most valuable thing in the universe, what I like to joke is the mirror of error said test. If you ask people what is what is the one thing you most want, and you really thought about it, most people would say a medicine for themselves or a family member. And so there is just immense intrinsic value to having one's own pipeline programs. And when we interact with pharmaceutical partners, on average, they love that we're working on molecules ourselves because it means that we're not just a bunch of keyboard jockeys, but we live the dream that we are pitching, and we need to use these models in practice for ourselves too. So they're battle-tested in an additional way, and we're sort of domesticated in that we know what's really difficult about the space. We're not just sort of selling some pipe dream.

It seems to me like there's been a shift in the last, I don't know, year to two years, where suddenly pharma has become interested in buying tools off the shelf. Whereas previously, there are a lot of companies that started trying to build AI and then had trouble, and so they became pharma companies because that's the only way they could make money, right? Is you raise a bunch of money and hope that in 10 years you have a drug, and if that drug is successful, then you've made it. And otherwise, then you fold or sell off the assets or whatever. And it seems like now that's changing. You have a lot of recent AI acquisitions in drug discovery and pathology and lots of different related areas. And so have you seen this shift? And is that part of the, are you shifting towards selling models, or is this just rebranding just because of that shift? Like, what's going on?

To finish that narrative arc there, the name change was just to reflect what we were really doing in practice, which is we are an AI company, as we have been from day one, when we were, you know, coding in PyTorch 0.17, building, you know, the first, when Sergey was scaling transformers on PyTorch, the first one to do that, we were scaling graph neural nets in PyTorch, one of the first ones to do that. So it's reflecting what our identity has been from day one. Um, but also alluding to the fact that we are very serious molecule makers ourselves. And we are full stack in that way. Um, we are also, to be clear, our mission is to create as many medicines as possible as would otherwise not have been created. And I think the way to get there is to put our technology directly into the hands of as many drug developers as possible, not just our own, but in big pharma, mid-cap pharma, and biotech. And that's partially why we've been building our agents, so that our models, which have become so much more powerful, can be easily used and orchestrated in the hands of med chemists, CAD scientists, who might have never ever written a line of code in their life.

I find it interesting to hear that you have now started bringing in agents into your development process. You know, it sounds like from the very get-go, your philosophy was much more, these are tools for scientists, and you make the best possible tool, and this lets them optimize whatever pipeline they're going to use, but ultimately, this is a tool for med chem. And now you are switching and you were saying, we are actually going to have automated systems which make decisions and then advance. First of all, you did allude to, you've kind of hit maybe the threshold to, you know, the like last November threshold where it just magically became useful. So maybe that's part of it. But are there other reasons for going in this direction? And like, you know, why now?

Well, let me ask you a question in return. How many tools do you think a human med chemist or CAD scientist can realistically learn and like be proficient in?

Oh, I was say, how many can they learn and how many do they use? Because I've seen that they like to use lots of them, but maybe they're not necessarily good with...

But it's really hard, right? A lot of the tools are extremely complex.

They have a lot of parameters that you need to set properly and configure them, right?

And this is where agents are extremely good at. Uh, they, they actually know how to use all of the tools and how to orchestrate them well.

Well, so I wouldn't interrupt you when you say agent, that some people like using the term agent to just mean something automated, but other people use agent to specifically mean you have an LLM which is responsible for orchestrating decisions. I mean, in outside of biotech, it's almost universally the latter, but sometimes in the space of bio, people are still using agents to refer to any sort of decision-based system which is automated out of humans, and maybe some of them are even more like traditional rule-based systems or something, right?

So in our case, I mean, AI definition of agents like an LLM which is able to use all of the tools that we have been painstakingly building over the many years of existence of this company. And that's a very powerful concept.

Because now, as a CAD scientist or med chemist, you don't need to deeply understand all of the details and all of the hyperparameters that you need to set, and you can just let this thing go and figure out and solve the problems. And this is where all of the previous questions that we discussed came back as very important. One, like agents need to be able to understand what these models are producing. Why do we care about crystal structure? Well, the agent can actually look at your predicted crystal structure and make decisions based on that. Something that we believe is super important to iterate and improve.

You're saying that you can take a crystal structure, which I guess in this case is framed as a graph or like a three-dimensional image, that Molestar is interpreting in some way, and then it actually can make decisions upon what is binding to what and make hypotheses? Like, you've gotten to that point where you have seen it making effective decisions?

Right. So there are definitely multiple layers to this question. You can start with like basic image understanding, right? Or you can use additional tools to understand what is happening in the crystal structure, or you can even train a model which is able to natively understand crystal structure representation. And like, image modality is a modality for LLMs these days. Crystal structure can be a modality for an LLM as well.

I can fine-tune it on sequences that look like that are somehow capturing tokens that are 3D crystal structures.

Right?

Yeah.

So, if you historically have treated humans like you were making tools for humans, so now you have agents, how do you reimagine the interaction of humans and agents?

And the...

decision-making process to go? Is this you hope to essentially automated humans out? You want to scale up and do more hypothesis or Yeah.

And then also the kind of the the the skeptic would ask, how do you also tell make sure that your agents are not making weird decisions where they tweaked the wrong hyperparameter and now you give you the human a bunch of >> Yeah. You don't you don't understand the parameters well enough to know that that happened. Yeah.

>> It's it's a great question and I think the answer here you can see in how coding tools have evolved, right? You remember Copilot when it just appeared and it was like top top autocomplete. >> It was still useful, but not to the extent that it is useful right now where I can fire up Copilot and it just goes and does everything for me. So I think we're going to see a very similar trajectory here. Now, those agents were already useful, but now humans need to provide consistent feedback and basically steer and guide this this agent along the way. But over time, we're going to become more independent. Still, I don't believe in full automation like replacing humans. I believe humans are going to become way more efficient in the job that they're doing using the tools. So a human will be providing strategic direction to what this agent needs to do, and the agent is going to be able to go and execute on behalf of a human.

>> I think this is the best time ever to be a creative human being, or really be any human because like the long a long arc of history starting from when we decided that hunting and gathering, I don't know about that. I think I'm going to settle down and and and do this whole agriculture thing. I think that started thousands of years of drudgery. Sometimes you wonder why did we actually do agriculture because everything I've seen seems like agriculture was probably a net loss on the short term.

>> It was the only benefit I see in the long run is that if you were a hunter-gatherer and you got sick or born with a genetic illness or got injured, that was like you're in trouble. Uh, there's not much more we could do for you. The medicine just didn't really exist. But ultimately, I think the crowning achievement in my mind to civilization is where the sick people among us in society are not seen as burdens, but people who we want to help. And that isn't true in the history of the animal kingdom, including humans. Um, it's a sad reality, but it's true. Whereas now, because of medicine, we we see people who are sick as people we want to help, who we can help, and we can help them. They could become even more productive members of society if that's the thing that you care about.

If to fast forward that to now, I think that so much of creative work, day-to-day work, if you look at as a pie chart, time is our most precious resource. So much of it was being taken up by repetitive tasks that actually didn't require much thought or creativity. And now it's like we've created like this additional headspace for deep thinking. And for us, you're solving some of the most technically challenging but but important problems. We can use these tools to become wildly more productive and creative. And that goes to the med chemists, drug hunters, CAD scientists who are our direct users, that they can be grand strategists for a drug discovery campaign. And when you have hundreds of drug discovery scientists working 24/7 on large data centers, just the output you get in terms of discovery is just going to be greater, especially if it's used in the hands of expert scientists who know what they're doing.

>> We probably should have talked about this earlier, but you uh have some really exciting results from your your Pearl model based upon essentially an a recent open uh data set challenge that you alluded to earlier. Um, sorry we didn't talk about it before, but I would like to give you a chance to talk about this. So there's this open bind was it EV A721A protease uh which was a very very hard target had a lot of things which made it hard for the community or for co-folding methods to internally work and uh you ran this internally and I think had some you know fun results.

Yeah, I mean, we can also go back to our conversation of on evals um and what's wrong with evals in general. Like if you look at the open source models and their performance on public benchmarks, it's kind of very similar across the board. Um, and what was surprising when OpenFold came up with this new benchmark is that you see a wide variety of performance of the models on that. And well, in part of it is because it's just a single target. Okay, I get it. But it's also because this is the target that those models haven't seen during the training. So nobody has optimized for that specific target before. Uh, nobody climbed on it. Um, so it was very interesting for us to see how our models is going to perform out of the box on this particular target that we also have never seen before during training or when we developed this models. Um, so when it came up, we obviously wanted to look into it. We we run Pearl Pearl system on that uh, and we were genuinely surprised and excited about results. Our numbers are way higher than um other published uh openly available models. And to me, it reflects how Pearl actually behaves in real drug discovery programs cuz we have seen this kind of number differences shifts in internal programs or partnership programs before, but we were unable to talk about them cuz a lot of this data is like it's either partnership data or internal data we cannot disclose it. Here you have a perfect example where it's an external target, but we didn't know um, and we just run on it. Um, and we see this staggering difference. Um, so that's kind of what makes me excited. And there are some specifics of this target like it has this flexible loop, but needs to move uh when you insert your ligand into the right location, and a lot of methods just are unable to handle it. And where Pearl was exceptionally good is figuring out how to move the loop, and we are basically correct for every single pose. Uh, like we're predicting really well uh this movement. Um, yeah, so that's the exciting part.

>> I'm curious about this. What is it about Pearl that allows it to model a dynamic target like that?

Pearl model the way it's trained, right? It's trained to predict together protein structure and ligand um, and like protein structure and ligand together, they're not necessarily in the same shape or form as the protein in isolation. So the whole training process it kind of nudges Pearl into that direction.

>> So are there lots of examples in the training set like that where the there's this dynamic nature where once when during the binding embed the confirmation changes?

There is induced fit throughout the training set. Yeah.

>> This is probably inherited from the the physics-based simulations that you're using.

>> It certainly helps. I think the other component that that Sergey alluded to was um, we haven't just been benchmark maxing on the same. We obviously publish and runs and poses. We published the state-of-the-art numbers and runs and poses last year when we published the Pearl technical report with Nvidia at GTC at the end of last year. We use runs and poses, we use pose busters. Pearl does the best on those benchmarks. We did that.

>> Um, however, as Siri is alluding to, the gap from Pearl to next best model is even larger on the uh partner drug targets that we work on. And um, what that's in part a reflection of is the objective of our company is not maxing benchmarks. The objective of the company is creating value for biotechs and pharma companies that are usually working on hard drug targets that are different from what's in the PDB. And so we've had to build the models in a way that are able to extrapolate. And the OpenFold challenge was just the latest example of that.

Before we wrap up, we always ask two questions. One is, if you could remove a bottleneck from your industry by fiat, right? So, and this is not like for inputs, outputs, uh, business, what would that be?

>> I can say where my main bottleneck right now is is in GPUs.

>> Okay. Everyone says.

>> Uh, yeah, look, GPU prices are going up and up and up, and LLM companies are kind of sucking up all of the GPU capacity out there.

And I firmly believe that what we're doing is genuinely important for humanity, like discovering new medicines for for yourself, for your loved ones, is it's something that we all need at some point in our life. Um, so I think it's very important to unblock that GPU bottleneck for drug discovery.

>> Is Anthropic actually throttling science because of GPU purchases?

>> Are you getting spicy?

>> I don't know if you can ask this, but are you looking at non-Nvidia? I mean, I guess you guys are Nvidia partners, whatever. So, but like there I've seen that some companies are pivoting towards a multi-architecture model so that you can take advantage of other suppliers because of the shortage.

Nvidia's been a great supporter of us. They've invested twice in Genesis, and we collaborate closely, like, you know, not just in terms of the capital, but we optimize kernels together. You know, we co- Pearl was co-authored by Genesis and Nvidia. They um publicized the OpenFold results with us. We've been very grateful for that. They're a very, very, very big company, and you know, we're we're not at the scale of a of a customer like a Meta or something. So we're grateful that they've given us time and attention nonetheless. But I think part of that is is a self-serving aspect here, which is that obviously, like no one can time time the market, but um, the amount of hype and investment in the traditional LLM space for chatbots, coding sort of thing, at some point, the amount of alpha will not be what's desired compared to the amount of investment. And I think regardless of the vicissitudes of the economy, medicines will always be high in demand. And I think whether it's a desire to help humanity or a self-serving interest in looking for the next the next big thing, I do think that chipmakers, including Nvidia, are going to want to get a lot more invested in life sciences because will always be high in demand. And the amount of alpha left in pure LLM space is just getting a little questionable. I mean, all the LLM companies are starting looking into life sciences right now anyway, right? They GTB, Roslin, there's a cloud initiative, there's lots of startups and whatever.

>> Yep. We think those are all great tailwinds for us.

>> Yeah. And then what what about you? What what's the bot on that besides GPUs?

>> Oh, I had the same answer as Sergey for sure. If I could wave a magic wand, I'd love to have just an enormous H100 issue cluster. Uh, that that would be that would be amazing.

And then the second question we ask is just, do you have a call to action for our audience? So meaning uh AI engineers plus scientists um get involved with something, come apply for jobs, uh do like what's the call to action for them?

>> We are definitely hiring. We're definitely looking for top talent in the space. And I personally, I transitioned from LLM space to drug discovery. I've never d done drug discovery before during Genesis. I find this field fascinating and extremely interesting. Um, you know, in LLM companies, you often find research scientists excited about working on architectures, but like kind of forced to work on boring stuff. And honestly, LLM architectures are relatively boring. I I don't know. I probably alienate half of your audience.

>> I think they probably think the same thing. But it's like it's a transformer layer in the end, like paper was published in 2017, and you go to any LLM lab today, you will see a very, very similar pieces uh in their models.

>> Yes, there is a little bit of flavor of MoEs, but otherwise architectures are fundamentally very similar to what was discovered back in 2017, right? Well, our architectures and our models are actually very different and very, very interesting to work with. So if somebody is excited excited about working on architectures, well, this space is actually a very, very interesting place to work.

>> Yeah, I like to comment. Not only are they very your architecture is very different from the LLM space, but you're actually you're also very different from what a lot of the other bio-ML space is doing as well. Like it's it's uh sort of unique both in their subdomain and more broadly.

>> We saw as a recent guest you had one of the uh the the progenitors of ESMFold.

>> Yeah. Um uh yeah, they cite our model architecture paper that we published and open-sourced a few months ago. So the work that you do here at Genesis as an AI researcher, there's lots of opportunities for publication, open source, and influencing the field beyond the obvious ways to influence the field by working with top customers that are actually making medicines that change people's lives. So, it's both incredibly intellectually engaging and a lot of opportunities to influence the field and and get your name out there. But if you want, you could also just stay being a cog in the machine of an LLM company. And

>> No shade, no shade. Yeah.

>> All right. Well, thank you so much for for coming and speaking with us. I've personally really enjoyed this. Hopefully the audience as well. Evan, Sergey, thank you. And we look forward to seeing how things develop. Um, and hopefully you get some good job applications out of this.

>> Yeah, we we love the show, so we really appreciate uh taking the time to to to chat.

>> Yeah, thank you for having us.