AlphaFold 3 Lands in Nature: Structure Prediction for All of Life's Molecules

Google DeepMind and Isomorphic Labs published AlphaFold 3 in Nature on May 8, 2024: a new diffusion-based architecture that predicts not just protein structures but their interactions with DNA, RNA, ligands and other molecules — covering nearly all of life's molecules.

Where AlphaFold 2 handled only proteins, the third generation pushes AI structure prediction from single molecules to molecular interactions, directly serving drug discovery. Isomorphic Labs has since advanced AI drug development and moved a pipeline toward clinical trials.

Later that year the AlphaFold team won the Nobel Prize in Chemistry — AlphaFold 3 is the latest chapter behind that honor, turning AI for Science from vision into productivity.

From 'Predicting Structure' to 'Discovering Drugs'

AlphaFold 2 solved 'what does a single protein look like,' but the real question in drug discovery is 'how do molecules interact.' AlphaFold 3 extends prediction to complexes of proteins with DNA, RNA and ligands — aimed squarely at pharma's core question of whether a candidate molecule can bind its target. This is not an accuracy-number bump but a leap from basic-science tool to industrial productivity; Isomorphic Labs' subsequent push of an AI drug pipeline toward the clinic (see our AI drug discovery coverage) is the direct expression of that leap.

Our Take

The AlphaFold series is the most convincing sample of AI for Science: it wins not on parameter scale or benchmark scores but by redefining how a basic scientific discipline works. Together with the same year's Nobel Prize (see our coverage), it marks AI's promotion in scientific discovery from 'assistive tool' to 'discovery engine.' The next thing to watch is clinical validation: however accurate structure prediction gets, it still needs lab and trial verification step by step — how much AI shortens the drug-discovery cycle will ultimately be answered by the number of approved drugs.