Google DeepMind published a paper today in Nature describing a significant advancement in predicting how proteins interact with each other and with small molecules. The new system, called AlphaFold 3.5, can model protein-protein and protein-drug interactions with 85% accuracy, a substantial improvement over the 60% accuracy of previous computational methods.
The breakthrough has immediate implications for drug discovery, where understanding how potential drug molecules bind to target proteins is a critical and time-consuming step. Pharmaceutical partners including Eli Lilly and Novartis, which have been collaborating with DeepMind, reported that the technology has already helped identify three promising drug candidates that are entering preclinical trials.
DeepMind CEO Demis Hassabis said AlphaFold 3.5 represents a fundamental shift in how drug development can be accelerated using AI. The company will make the interaction prediction tool freely available to academic researchers through its existing AlphaFold database, which has already been used by over 2 million scientists worldwide. Commercial pharmaceutical applications will be licensed through a partnership with Isomorphic Labs, DeepMind's drug discovery spinoff.
