DeepMind's AlphaFold 4 Accelerates Drug Discovery Pipeline

Google DeepMind has unveiled AlphaFold 4, an evolution of its protein-structure predictor that now models small-molecule binding affinity - the property that determines whether a candidate drug will actually work.

From Structure to Function

AlphaFold 3 predicted protein structures with atomic accuracy. AlphaFold 4 goes further, simulating how candidate molecules dock into those structures and estimating binding free energy, a metric that traditionally requires months of wet-lab work.

"We have compressed the hardest part of early discovery - going from a target to a viable lead - from 18 months to roughly three weeks," said Pushmeet Kohli of DeepMind.

Real Candidates, Real Trials

Three drug candidates discovered using AlphaFold 4 have entered preclinical trials through Isomorphic Labs and Eli Lilly, targeting cancer and autoimmune conditions. Prediction must still be validated experimentally, but the hit-rate improvement means far fewer dead-end molecules reach the lab.

The model is available to academics through a free tier, with commercial use licensed via Isomorphic Labs.