Valence AI on biology's 10^60 molecule search space for perturbation prediction

dom_beaini · x · 2026-09-18

Valence AI lays out its approach to decoding biology: understanding how genetic or molecular interventions alter cellular states. The challenge is scale—there are 10^60 drug-like molecules across many cellular contexts, an intractable search space even for the most advanced labs. Yet training perturbation prediction models that generalize requires sufficient data diversity. This is the setup of a thread, ending with the question of how to resolve the tension.

Original post →

More from Research

Research channel →