Google DeepMind launches AlphaGenome 2 for single-cell DNA prediction

Google DeepMind has launched AlphaGenome 2, a new DNA model that predicts how DNA changes affect specific human cell types. It helps researchers find the cell-level drivers of diseases. The model was trained on over 100 million single-cell maps from 68 tissue types, turning static cell maps into a prediction engine for genetics.

Unlike the first version, it zooms in from broad tissues down to single cell types and states, with a rebuilt architecture. In tests across 28 immune cell types, it spotted 30% of causal variants and judged gene effect directions with 43% accuracy, nearly double the first model. It also adds new DNA methylation prediction.

The model is now open to non-commercial research via the AlphaGenome API, while commercial access is in private preview. Google notes it is for research only and has not been validated or approved for clinical or diagnostic use.