🔬 The Coolest Diffusion Research Isn't in LLMs — Evan Feinberg & Sergey Edunov, Genesis Molecular AI

“

Evan“people love 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 disease, where the molecule is predicted to have good pharmacokinetics, as in the levels in the blood in patients or 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 because we often know what genes, what proteins, what targets are causing disease. They're just really hard to drug, or the molecules we put in the, in patients are inferior to 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 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 looking for more selective therapies. And so our view is that is the highest leverage application of AI in healthcare, medicine more broadly.”