A Genome Model That Reads Cell by Cell

AlphaGenome 2 narrows variant prediction from tissue to individual cell state, nearly doubling detectable effects.

Google DeepMind released AlphaGenome 2 on 11 November 2026, trained on more than 100 million individual cell profiles across 68 tissues, narrowing prediction from broad tissue categories to single cell types and states. The model roughly doubles the number of variants it can identify as altering gene expression compared with the first version, and is available to non-commercial researchers on application. The scientific value lies in turning a static cell atlas into a variant-effect prediction engine, which gives drug-target discovery and rare-disease research a new piece of AI infrastructure; Nature reported researchers describing the results as notable. Expect a run of validation studies from academic groups over the coming weeks.