AlphaProteo, announced by Google DeepMind in September 2024, is an AI system for designing novel protein binders, proteins built from scratch to attach tightly to a chosen target molecule. Where AlphaFold predicts the shape of proteins that already exist, AlphaProteo runs in the opposite direction, generating new proteins to perform a specified job.
Trained on the Protein Data Bank and on structures from AlphaFold, the system designs binders for targets such as viral and cancer-related proteins. DeepMind reported that, across the targets tested, its binders bound between roughly 3 and 300 times more strongly than the best previous methods, and that it produced the first AI-designed binder for VEGF-A, a protein involved in cancer and diabetes complications. The company also noted limits: AlphaProteo failed to design a binder for one difficult autoimmune target, TNF-alpha. The Francis Crick Institute helped validate results experimentally, including binders that blocked SARS-CoV-2 from infecting cells.
Strong, designable binders are valuable because they are the basis of many drugs, diagnostics, and research reagents. Designing them well has traditionally required testing huge numbers of candidates.
For a general reader, AlphaProteo extends the AlphaFold story from reading biology to writing it, and signals DeepMind’s push, through its drug-discovery arm Isomorphic Labs, to turn protein AI into medicines.