Robotic labs could bring drug optimization back to the U.S.

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 48:45

   I mean, almost all of them had big parts of their programs that happened in China. It very much rhymes with what we did with the manufacturing industry, where in the early days, it's like a bunch of sensible local decisions where you're like, oh, well, this is low-skilled labor. It can happen more cheaply elsewhere, so we'll move it out. And then over time, it becomes not just low-skilled labor, but, you know, there's innovation that happens and suddenly our muscles atrophy. There's grow and all of a sudden we can't manufacture a lot of the things we need. There's this amazing quote from Andy Grove from like years ago when he was running Intel. He said, abandoning today's commodity manufacturing can lock you out of tomorrow's emerging industry. It happened exactly in manufacturing and it is on the way to

2. 49:31

   happening in pharma and we can't let it. So with, with manufacturing, I remember the story of American de-industrialization as driven by two factors. One is what you said, lower labor costs abroad, but two is it's dirty, nimbyism. I don't want the smoky steel mill in my backyard. Let's put it halfway across the world where I don't have to see it or worry about it or think about the pollution. With CROs, is nimbyism a factor whatsoever? I feel like no one would care, but how has that played into it or has it been a non-factor?

3. 50:08

   I don't think that's a huge factor here. I think it's been much more of the first that you identified. It's the cost and it's this feeling of, oh, well, maybe this won't be so strategic. And then, of course, it turns out that it's massively strategic. We don't want to lose the ability to turn ideas into medicines here in the United States. Yeah. So then on the labor cost, it feels like by leaning into higher volume tests, testing driven by robotics, you can be more competitive on the economic side. So I would imagine that within, I don't know what your timeline is, a year or two, like on a cost per experiment, you would be competitive internationally very quickly, correct?

4. 50:46

   yeah and so the like just to step back because i bet a bunch of people don't know how these processes work so let's say the three of us go start a biotech company we're going to develop new medicines we're going to try and cure a particular form of cancer you usually start with some biological target right we might say it's usually a protein so we might say you know maybe because of jordy's grad school research we've got uh you know you either want to like yeah you either want to like uh you either want to accelerate or inhibit some protein Sure. And and you've got that idea. You find a molecule that that maybe sort of does it, but it probably is toxic or it's not soluble.

5. 51:22

   So the body can't process it the right way or it's not potent. So instead of taking, you know, a tiny amount of it as a pill, you have to take a kilogram of it, which, you know, all of these things disqualify it as a drug. And so the next step in the process is we would go through three to five years of making variations of that molecule to try to optimize it, to make it more potent. to make it more soluble, to make it less toxic.

6. 51:49

   So there's this like three to five year process and that entire process is outsourced to China. And that's the problem. And that's what we're bringing back. And the only way you're going to bring it back to your point about economics is if you do it a different way. You've got to use AI and robotics and that's what we're doing. Talk about like how you see demand over the next couple of years. I can imagine in the same way that we're seeing an explosion of new cybersecurity vulnerabilities, We might see an explosion of new, like pretty promising, even like concepts or ideas for new drugs.

7. 52:22

   Do you imagine you're going to very quickly be, you know, even as you're expanding capacity, like turning away a lot of opportunities just because of how many opportunities there are? Yeah, I think this is the this is the exciting thing about where we are. This is a real existing industry. Billions of dollars are going right now overseas to these players to help develop every drug that you ever see or take.

8. 52:53

   And so there's already a big market for it. And every player would prefer to have that done in the U.S. where they have strong IP guarantees and everything else. And to your point, you've got lots of exciting new players in AI. Every one of the frontier labs is getting involved, which I think is amazing. And you've got all kinds of new startups that are using AI to do bio. I think that's only going to accelerate.

9. 53:15

   So there's not just a big market today. There's a big and growing market going forward. On the robotics side, what is the opportunity to pull stuff off the shelf? I'm assuming there's a ton of companies working specifically on that. and you can sort of acquire existing robotic tools or I imagine a big part of this will be building your own, but what do you think your view is? Yeah, there'll be some of each.

10. 53:43

    There is a lot happening. I mean, this is one of the reasons why, this is like the why now for this company is robotics is just at the beginning of the S-curve that we've seen AI, you know, more broadly go on. And so like the capabilities are increasing at an incredible rate. For us, we're going to be super pragmatic. If there's something that we can build, that we can buy, that helps us go faster, helps us automate more of the process, then for sure we'll do it.

11. 54:08

    But I don't think that it's all out there ready to be bought. I think this is going to have to be a build from the ground up, which is why we are totally vertically integrated. We have chemists on the team. We have roboticists. We have mechanical engineers. We have AI researchers. And I think that's the only way that this company is going to be successful. Can you talk about the shape of that S-curve?

12. 54:27

    My intuition is that we saw sort of a fast takeoff in software vulnerabilities because you can run PHP in a virtual machine and you can hammer it with a ton of different exploits that you're inferencing pretty quickly, inferences pretty fast, and everything can be done in this closed loop that happens so, so quickly. And right now in bio, maybe it's a week to run an experiment, get it back. Maybe you bring that down to a day, maybe an hour.

13. 54:56

    But even if there's a few minutes of timeline, you're looking at an order of magnitude slower than software. So I would assume a little bit slower of a takeoff. But am I wrong about my intuition there? I think this is exactly why, even as we're seeing AI completely remake software engineering, and you're seeing AI do things like solve Navier-Stokes and make progress on the Riemann hypothesis, we haven't seen the same level of of sort of mastery in the physical sciences in chemistry and biology and material science it's exactly the reason that you say yeah and that's one of the reasons that it's so important that we develop robotic labs right it's it's both we need to restore this from china and this is the way that you generate data at scale All right.

14. 55:46

    that will allow us to build the next generation of incredible scientific AI models, right? And because there's nothing, I can't think of very many things more important than seeing that kind of acceleration that we've seen in software and math, but in biology and chemistry and drug discovery instead. And it might be multiplicative. Like if you shorten the time for the loop and then you're applying the models faster, you actually get an exponential there that you see results very quickly.

15. 56:13

    Exactly. I mean, the thing that's going to come when AI is able to develop and to design more drugs is then we're going to need to test way more drugs. And that will, I think, invariably require a physical component. But that physical component also generates really valuable data. And so there's absolutely a loop here.