LONDON — Google DeepMind has released AlphaFold 3, the latest iteration of its protein structure prediction AI, and the leap forward is staggering. Where AlphaFold 2 predicted the static 3D structure of individual proteins — a Nobel Prize-winning achievement — AlphaFold 3 models proteins in complex with DNA, RNA, small molecules, and potential drug compounds. It has mapped all 200 million known proteins in dynamic interaction states, essentially creating a comprehensive simulation layer for molecular biology.

The pharmaceutical industry has responded with unprecedented speed. Novartis, Pfizer, and Roche have each announced partnerships to integrate AlphaFold 3 into their drug discovery pipelines, with initial applications focused on previously "undruggable" targets: proteins implicated in cancer, Alzheimer's, and antibiotic-resistant bacterial infections whose structures had resisted experimental determination. Early results from Novartis suggest the model can reduce the time required to identify a lead drug candidate from 18 months to as little as 6 weeks.

DeepMind has released AlphaFold 3 as an open-source model under a permissive license, a decision that CEO Demis Hassabis described as "a moral imperative given the potential to accelerate treatments for diseases that disproportionately affect the developing world." The academic research community has embraced the tool: over 2 million researchers accessed the AlphaFold 3 server in its first month.