The Quest to ‘Solve All Diseases’ with AI: Isomorphic Labs’ Max Jaderberg

0:00We have set up the company from day one to really go after this big ambition. This isn't about developing therapeutics for a particular indication or particular target. It's really thinking about how to recreate a very general drug design engine with AI, something that we can apply to not just a single target or even a single modality, but we can apply this again and again across any different disease area. And that's what we're stepping towards at the moment.

0:48Today we're excited to welcome Max Yoderberg to the show, Chief AI Officer of Ice and Morphic Labs, which launched out of deep mind with a goal of revolutionizing drug discovery using AI. Last summer, they released AlphaFulp 3, a stunning breakthrough that allows us to model not just the structure of proteins, but of all molecules and their interactions with each other. That led to Demisisobis winning the Nobel Prize in Chemistry last year. Max describes their vision for what a holy grail model for drug design and what agents for science look like. He draws parallels to his experiences building alpha star and capture the flag and the research directions of building agents and games more broadly.

1:30Specifically, with 10 to the power of 60 possible drug molecule structures, we need to build both generative models and agents that can learn how to explore it and search through the whole potential design space. Max also describes his vision for what a GPT -3 moment for the field might look like, describing it more akin to AlphaGhost's famous Move37 when we start to see things that exhibit superhuman levels of creativity in AI drug design. And that's done even humans ourselves. This is one of my favorite episodes yet. Enjoy the show. Max, thank you so much for joining us today here in London.

2:08Nice of pleasure to be with you here. Yes, fantastic. Awesome timing to with the launch of Alpha full three and with Dennis winning the Nobel Prize in chemistry, which is a true testament to everything that you and your team have done over the last couple of years. Yeah, 2024 was definitely a busy year for us. Lots of big breakthroughs. No surprise was just incredible to see, you know, I think amazing recognition for this, for this seminal piece of work. Yes. Well, I'd love to start with talking a little bit about your own personal story. You've had an incredible career in the world of deep learning from the very start.

2:42Authoring many seminal papers while at deep mind, including for capture the flag and alpha star. breakthroughs in the world of deep learning. Can you walk us through some of the key questions that you had in your field of research around reinforcement learning at the time? Yes, so at deep mind, I worked on a whole host of stuff, early days of computer vision and deep generative models, but it was really reinforcement learning that ended up hooking me there. Deep mind was the place in the world to be working on reinforcement learning at that time. And really the question in our minds was, how can we actually get to a point where we could get an AI that could go off and do any task you wanted it to do?

3:32And the dominant paradigm at that point in time was supervised learning. And supervised learning is very different from reinforcement learning. They're both learning techniques, but supervised learning. you need to know what the answer to your question is and that's how you train the model. So in Supervisor Learning you give an example and then you supply the model with the answer to that question. Now that can be great if you already know everything about the problem that you're training this AI to do, there's no network to do. For most times you don't. Yeah I mean there's just so many problems in the world where we don't know what the answer is, we don't know what the solution is, and if you think about, you know, I think about how I want AI to be applied to the world, yes, it's going to be great to be able to apply things where we're already good as humans here, but really, you know, the big frontier is can we start applying AI to places where, you know, humans don't know how to do this stuff or, you know, there's a limit to human performance there.

4:36And, you know, that's where reinforcement learning is one of the key tools and has real promise here, because in reinforcement learning, you don't need to know what the answer to the question is. You just need to be able to say whether the answer that the model gave you was good or not good. Maybe even how good or not good. And so this opens up a completely new, field of problems to train these models against. And so reinforcement learning and really starting from what was one of the big breakthroughs of deep -mind in the early days was working on games like Atari. The question was, okay, so how can we scale this up from the world of Pong and space invaders to things that really start to look like real problems in the world?

5:27And so there was an amazing track of research as we scaled up these methods. Yeah. Did you know that Sequoia was the first investor in Atari back in the day? I really. I didn't know that. That's incredible. Yeah. Yeah. No, there's Atari games where, you know, great fun actually to, um, to sort of go back and play in the context of, hey, we've got an agent and, you know, I'm just gonna have a game of pong on the side as well. Yeah. There's a wonderful wall at Sequoia in our office where we've all these names of legendary IPOs and M &A's that have happened And there's one, I think it's called the pizza company.

6:06And I love asking folks if they know what that is. And it's actually from Chuck E. Jesus, which was an original Sequoia investment at the time. Amazing. Amazing. So, capture the flag in Alpha Star were incredible breakthroughs at the time. Can you share a little bit about what exactly those breakthroughs were? And maybe why you chose those specific gains? Yeah, so, you know, If you think about the history of AI using video games, why do we use video games at all? Video games are these sort of malleable, perfectly encapsulated worlds that as researchers and scientists, we can manipulate them. We can test out different algorithms in the, we can set up different situations.

6:50So the perfect test ground for us to develop new algorithms. And then you can imagine as a RRL researcher as someone who's like, thinking about how can we get AI to be as general as possible. You're always thinking, okay, we've cracked Atari, how do we get a more complex game? Yeah. And the thing that I was personally obsessed with is I want these agents to be able to zero -shot, be able to do any task. And this is a slightly different paradigm from what people were doing at the time with training on Atari, where normally when you're in enforcement only you think about, here's a game, now you get to train on it and get good at it.

7:34And then you apply that same algorithm from scratch training on different games. Yes. I'd love a different scenario where instead we train an agent and then we can lift it and put it on any new task and that agent will be able to perform well in that task without any more training. And so to do that, what you're really asking for is generalization over task space. Yes. And that means you need lots and lots of training tasks. So the training data in this RL for agents becomes tasks. Not images, not piece of text, but tasks. And so you can imagine you could go and sit and take a whole game studio and try and hand -author hundreds of different tasks It's not for the mini games in these virtual worlds.

8:23And we did that. We were doing lots of that. And then you can think, yeah, we can actually go further than hand -awthoring. We can procedurally generate these tasks and games, generating worlds and maps and different objectives. And we did that. But you keep running into this complexity ceiling. that there's only so much complexity that you can hand -auth or you can design humanly. But that's where multiplayer games come in. Because as soon as you go from single player to multiplayer, it's not just the agent playing. You've got another player in this game and that other player or other players can take on many different characteristics and many different behaviors.

9:10So every different player, every different strategy that you're up against, changes fundamentally the game and what the agent is trying to do. I go back and think, why are people still obsessed with playing chess? Why does a professional chess player still keep playing chess the same game? But it's actually not because you're playing completely different opponents day after day and new people into the world. So the game is continually changing. So multiplayer player games and multi -agent games really encapsulates that huge diversity of tasks that you might encounter just from other players being there.

9:48And so, Capture the Flag was actually one of our first forays into how can we use multiplayer games to really stretch what our reinforcement and any algorithms can do to really force us to think strongly about how we can generalize to new tasks, how we deal with these multi -agent dynamics. So Capture the Flag was a fantastic breakthrough, really showed that we could get to human level performance for these multiplayer first -person games. And then of course, Starcraft added on a huge amount of complexity and was sort of the next frontier that we had to go after for this. You were so early in this that so many of these concepts are very, very relevant today in the world of language.

10:30How does it feel to see some of this were it continue to be played out? Yes, brilliant. It's just fantastic actually. There were so many things that we were talking about at that time. Seven years ago, yeah. Yeah, yeah, yeah. 2015, 16, 17, 18. And to see all of these core fundamental concepts being really useful and really applicable today in the world of large language models. And resulting in performance that we could only really dream about at the time, that's incredibly of resatisfying. So then, in your own words, you said that you moved from building toys to then finding real applications.

11:11When did you know that you found the right recipe?

11:16So I just love deep learning. I've been obsessed with deep learning for, you know, 10, 15 years now. And the thing that I love about it is that you have these underlying core concepts, these fundamental building blocks that are somehow incredibly transferable between different application spaces. So it's the same building blocks that we were using in computer vision in 2012 as we were using in, you know, the generative models in language, you know, then reinforcement and et cetera, et cetera. So what I was seeing just again and again was this ability to take these core concepts, these same core concepts, take incredible people who understand how they're almost like master chefs of putting these concepts together and these different building box together, take a team of incredible people and go after really, really challenging problems, You know, problems that you go to conferences at a time and you talk to leading researchers in the field, they say, no, no, no, this is 10 years away.

12:29And in the back of your mind, you know, okay, we're basically cracked here. Wow. And I saw that happen again and again and again. You know, you take amazing people, amazing algorithms, amazing computes on really challenging problems and we can find recipes now to crack so many problems. and so it just got to the point where, and I've always been quite obsessed with the application of these methods, I want to see this technology have, you know, real transformative positive impacts in the world. And so, you know, we need to start actually going after that. And the time has been right for I think a few years now.

13:10Yeah. Well, so you've now had a decade -long relationship working together with one of the greatest scientists, technologists, and founders of our lifetime, Demis. He called you while you were still at Oxford. And then your company, Vision Factory, and DeepMine, were both acquired by Google back in 2014 around the same time. And that's when the two of you started to work together now for over 10 years. What was it like or what has it been like to work with Demis? Yeah, I mean Demis is an incredible person, you know, a real character and a real visionary. And you know, also amazingly human and relatable and I think that really inspires people.

13:58So it only takes five -minute conversation for him to really bleed out the depth of ambition that he thinks about. And just the immediacy of the potential to get, to step towards these ambitions. So I think he has this great ability to inject a lot of energy into a group of very smart people, get people to see beyond what's right in front of them. I remember moments sitting, we're standing in the lobby of one of the early deep -mind offices. I think this was the, it was a toast. We were a celebration we were having for the first nature paper from deep -mind. Wow. And Dennis was saying, you know, this is actually just going to be the first of dozens of nature papers.

14:57And at the time, this was basically the first machine learning paper in nature. This was the Atari DQN paper. And the prospect of dozens of nature papers, you know, it seems very far fetched. And actually, he went further and said, and we're going to be winning, though, what prizes as result of this. And there was ten years ago. Yeah, that's incredible. But the fourth thought that he has, he's got what I call one of these roll -outs minds, maybe comes from all of his experience playing chess, but he's always rolling out into the future. What are the steps now that are going to lead to this big ambition?

15:34So yeah, it's been fantastic. I've been working with him for about 10 years now. Still work really closely together, nice and more for labs, and the ambition is as big as ever. It's so interesting to hear that you had this ambition and that he had this ambition from the very start and it's incredible that it's played out that way. Well I'd like to talk a little bit about isomorphic. You're now embarking on one of the most ambitious missions of our generation to rematch and drug discovery and drug development with AI. If everything goes right and you realize your vision for isomorphic, what does the world look like?

16:12Yeah, you know, we think really big isomorphic. We want to be solving all diseases here and genuinely that scale. And the point is that this technology that we're building, you know, and AI as a whole field is going to be completely transformative in how we understand biology in our ability to manipulate and craft chemistry to modulate that biology. So we really think about a future where we are solving all diseases where you know AI is not just helping us you know discover and create and design new therapeutics but also just understands so much more about our biological world about how our you know cells are working and what are the root causes of of disease and therefore opening up new pathways that we can think about modulating.

17:09So we we have set up the company from day one to really go after this big ambition. This isn't about developing therapeutics for a particular indication or particular target. It's really thinking about how do we create a very general drug design engine with AI, something that we can apply to not just a single target or even a single modality, but we can apply this again and again across any different disease area. And that's what we're stepping towards in the moment. How does setting out with this ambition of being general change how you built in practice from day one? Yes, a good question. When I think about some of the status quo of AI and drug design, there's been a lot of use of machine learning models in chemistry and biology, but I would call them a lot of the first generation of this sort of application to be more local models.

18:15You might have some data about a particular target or about how particular class of molecules is behaving and you'll fit a small, you know, multi -layer MLP against this data to help you generate some predictions that lead to your next round of design. This is the complete opposite approach of what we were trying to do. So from day one, we were setting out to create models that generalize across chemistry and across target space. So, you know, and a key example of this is something like Alpha Fold, Alpha Fold 3, where this is a model that you can apply to a whole different host of targets. You can apply to any protein in the proteome in the universe of proteins.

19:05You can apply it to any small molecule that you can think of designing without needing to find tuna, without needing to fit any local data. So you can imagine that it just completely changes the way that chemists can use these models if you don't need to be adapting this model to every single application. So every single one of our internal research projects, and by the way, when I think about what we're going to need to get this breakthrough drug design engine that we've been building, we need like half a dozen half a folds. Half a fold is just part of the story. So from day one, we've been setting up these internal research programs, going after these half -dozen problems.

19:49We've had significant breakthroughs, obviously in alpha -folds and structure prediction, but also in other key areas.

19:57And in all of these models are general. They can be applied to any target. And then what we're finding, actually, they can be applied to any modality, or lots of different modalities. Yeah. So that's the first time I've heard you say, half a dozen alpha folds. Can you share a little bit more about what that means? Yeah, alpha fold was obviously a massive breakthrough in understanding biomolecular structure. So what is the structure of proteins and now with alpha three structure, processes with small molecules and things like DNA and RNA? That's a fundamental step change. It allows us to get experimental level accuracy of a really core concept of biochemistry that unlocks a whole bunch of thinking and design work for chemists.

20:43But you know my comment here is actually we're probably going to need something like half a dozen more of these sort of breakthroughs. They're sort of getting to experimental level accuracy of different core concepts of biology and chemistry. To be able to put this together into something that's really transformative for drug design. Drug design is really, really hard. It's not just a single problem. It's not just about understanding the structure of a protein. Yes. It's not even just about designing a molecule that will modulate that protein in the way that you want. You want this molecule to be able to ideally be taken as a pill and go through the body and be absorbed in the right way and reach the right cell type and actually go into the cell and and not be broken down by the liver in a certain way.

21:30So there's just so much complexity to hold on to as a drug designer. And each one of those is like you know alpha -fold level style breakthrough that we've been creating. So interesting. Well I've also heard you use the words a holy grail model for drug design and agents for science. Can you explain a little bit more about what you mean? Yeah so some of these research areas that we've been going after predicting Structuring, and properties of these molecules and how all of these like biomolecules interacts and play out over time And these really are sort of holy grail predictive problems for drug design And we've made some incredible breakthroughs there which have you know really stunned our chemists and and step changed how we do drug design internally I say But what's I think a really interesting thing to think about is that you could create the best possible predictive model of the world, like an experimental level, even better than experimental level model, to predict a particular property about a molecule, for example, to be able to predict the outcome of a real experiment.

22:41See, we could have a whole suite of those, but that still wouldn't solve drug design. And the way to think about this is, there's this number 10 to the power of 60, which is perhaps all of the possible drug -like molecules that could exist. That's maybe, that's maybe, you know, takes into account a lot of things. So we could even reduce that by 20 orders of magnitude, get to 10 to the 40. That's still a lot of things. Yeah. And even if you had the best predictive models in the world, so let's say you could screen a billion different molecules, you could go and test a billion different molecules.

23:24That's 10 to the 9. Yeah. Now we're still like 10 to the 31 molecules left on the table. Yeah. So even with the best predictive models, you're still not even scratching the surface of molecular space that you should be exploring. And this is why we need to go beyond just predictive models of experiment, but also models like generative models, like agents that can actually navigate that whole 10 to the 40, 10 to the 60 space. That's so interesting. Using our predictive models obviously to understand how to navigate that, but so we don't have to exhaustively search, because we can never exhaustively search the whole universe of molecules.

24:03If that makes sense, just in the same way that, an hour ago couldn't exhaustively search all of the possible go moves, unlike chess, where you could exhaustively search all possible chess moves. Yeah, yeah. But yeah, molecule designs much more like go than it is like chess. So that's where generative models come into play, agents that utilize generative models, utilize search techniques as well as these amazing predictive capabilities to really open up the entirety of molecular space. Now, to me, it's actually still amazing that even without AI, we managed to find drugs in this 10 to 60 space, 10 to the 40 space.

24:44It just says that actually there's probably a lot of redundancy, there's a lot of potential designs. If you think about a particular disease and a location, a particular target, there should be many designs that exist that would be good for that and would be the right sort of profile for this therapeutic. And I think the real potential here is for these gerative models, these agents as well to be able to search through this space and really uncover that whole potential design space. That's so interesting. I think in very simplistic Leamann terms you're both modeling, learning and modeling the game and trying to build the best player to solve different types of games.

25:28Yeah, so I mean, you know, I'm incredibly biased by games. I've been playing video games as a kid. Grouper in that world. But you know, that's exactly how I think about it. We've got to be creating our world models, our models of the biochemical world, our biological world. And then we don't stop there. We actually then need to be creating agents and and generative models that can work out how to explore, how to traverse that, and to basically uncover these amazing needles in the hay stack of chemical space, which could be life -changing therapeutics for so many millions of people. I love that.

26:07That is our punchline. That's it. So alpha -poll -3 is truly groundbreaking. You've taken us from being able to model just the structure of a protein, to now being able to model the structure of all molecules, and their interactions with each other. Can you share a little bit about how we should think about that in terms of the impact and accuracy in speed and efficiency and also potentially in being able to explore problem spaces that we couldn't solve before this? Yeah, so alpha -fault 2 was the biggest breakthrough. To be able to understand the structure of proteins and then there was something called our fold to Multimer, which then allows you to understand not just the structure of proteins by themselves, each individual protein, but the structure of proteins is they come together and what we call complexes, so how these proteins fit together.

27:03That opens up, you know, and helps us answer a lot of questions in biology, but there's still a big hop to designing therapeutics. And one of the big classes of therapeutics is what's called small molecules. So these are molecules that are not proteins. These would be things like caffeine or parasitamol, things that often more often you can take as a pill. And the way that these therapeutics work with these small molecules is that they go through the body, they go into the cell. And they actually come and attach themselves to these proteins. You know, these proteins that are the fundamental building blocks of life, they form these molecular machines by interacting with other proteins.

27:44And so you can imagine that if you have another molecule, your drug that comes in and attaches itself to a protein over here, then it might disrupt the ability for that protein to interact with another protein, one of its normal machine and day -to -day life. And so you're modulating the function of that protein with this small molecule. And that's the essence of drug design and how therapeutics work. And so you can imagine as a chemist, your job, a drug designer, you're trying to design a small molecule that's going to fit to this protein over here and disrupt how it normally functions or in some cases enhance how it normally functions.

28:24And so it'd be really helpful to understand how this small molecule interacts with the protein. What's the structure that it might make? What are the interactions, these literally physical interactions that are being made? And so that really inspired the creation of Alpha Fold 3, where now we have a model that not only predicts the structure of proteins, but how these proteins interact with small molecules, also other fundamental molecular machine building blocks, things like DNA and RNA. And this basically opens up the ability to structurally understand, which is a core part a drug design. Small molecules, it opens up new classes of targets.

29:10There are things like transcription factors which are proteins that sit on DNA and read DNA and you can imagine now trying to design a small molecule to change or disrupt the function of something like that. To do that, you'd really want to be able to see literally in 3D how this all looks. And if I make changes to my little molecule, how will that change the way it interacts with this protein and this biomelectic system? So Alpha4 -3 is now very, very accurate. It allows us to answer a lot of these questions purely in silicae or purely in the computer where before you would have to go to the lab, literally crystallize this stuff.