“A rat isn't actually a human. It's not actually very predictive … if one of my AI scientists or engineers came up to me and said, hey, we've got this amazing model, it's 30% predictive, you're like, okay, maybe do a bit better than that. And so I think there's a lot of opportunity to actually rethink what those predictive endpoints for human clinical toxicity are. Forget animals. They've been a proxy for the last 50, 70 years. But we can maybe do better.”
sputnikintelligence.com
Elad Gil asked how Isomorphic Labs can predict side effects when animal data often fails to translate to humans. Jaderberg, Isomorphic's president, points out that rat toxicity studies predict human toxicity only about 30% of the time, and proposes replacing them with AI and lab-based methods.
…if you're trying to make a general solution to drug discovery. If you really care about generalization out of these models, there's maybe a different distribution of data you need. So we go out and generate this ourselves. And we've seen some really, really amazing data scaling laws as a result.
…translate into people. And so a lot of the data you actually needed, you had to generate in human clinical trials. How do you go about generating that data that actually informs it? Because if you're using animal models, often the data that comes out of that is incorrect relative to human biology.
…It's very painful. It's very painful. And if you think about toxicity, like sources of toxicity, like the gold standards that you need to get through for regulators is you need to prove that your molecules aren't toxic on rats and other animals. But still, even when things are approved and they've been proved on rats, they end up being often toxic in humans. And why is that? Well, okay, a rat isn't actually a human. It's not actually very predictive. I think, you know, choose your study, but it's about 30% predictive of human clinical talks in a rat. And if one of my AI scientists or engineers came up to me and said, hey, we've got this amazing model. It's 30% predictive. You're like, okay, maybe do a bit better than that. And so, I think there's a lot of opportunity to actually rethink how we, you know, like, what are those predictive…
And for these sorts of approaches or techniques, do you tend to partner for them? Do you bring them in-house? I'm a little bit curious as a company how you operate relative to, because there's so much that could be done.
Yeah.
And a little bit curious how much you want to just emphasize the computer science side. And then, you know, biology, often there's these things called CROs or these contract research organizations that people use. Unfortunately,