AI drug discovery needs evidence from human biology

Transcript

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

1. 12:39

   society. And so what are you working on that side of things in terms of the trials and the development? Yeah. I mean, I think this is the key thing that I think is very difficult in drug development. Everyone says like, yeah, like these models can like design all these drugs and things like this. But, you know, the problem obviously is not, you know, design the drugs, but actually getting into humans and i think also the challenge is like yeah i think the challenge in for drug development specifically is that um you have to like basically like let observe human biology and see how it works in human biology because the problem is a lot of this stuff is being done you know of course it starts out in silico uh kind of like you know with the simulations and with ai designing it and then it goes it moves into in vitro uh where it's just like in a test

2. 13:25

   T-borne in the lab, and then finally it goes preclinical, where it goes into mice and things like this. And all of these sorts of scenarios aren't very representative of what's actually happening in human biology, right? What's actually happening in the human body. And so you may see stuff that works really well, like in rats, for example, rats and mice. You can basically cure cancer maybe in rats and mice, but those drugs may not generalize well to human patients. And so that's why I think like we need to actually be training AI models on clinical data and on human biology. And I think that's something that is kind of underappreciated right now. There's a lot of work in AI for biology, but I think there needs to be more work in training these models on clinical data and understanding human biology with these foundation models. And then that can help accelerate

3. 14:12

   a lot of what I think is one of the main limiting factors is actually understanding whether or not these drugs will work directly in the human in the human patients and that's something you guys are directly working on as well yeah i mean i think like this the sort of directions that we're taking with things like holland s1 can help us understand um you know for example you know you're looking at you know um the the drug may for example um work in one set of patients but they'll work another set of patients and holland s1 can show well you know there's maybe some specific biological pathway that's more active in the patients that aren't responding to the drug and so that's something that, you know, then you can understand, like, what's actually going on in the human biology. And maybe that's something that also needs to be targeted, for example. And so that

4. 14:54

   also kind of brings you to, you know, discovering new treatment targets and for discovery. But more importantly, it's actually grounded in the biology of the human. And I think that's what's the real sort of kind of promise of or potential of these sorts of tools is, you know, actually grounding what we're doing in the human biology. Yeah, definitely. What's pretty interesting is in the