Google DeepMind launches AlphaGenome 2 to predict DNA effects on cells

Google DeepMind has unveiled AlphaGenome 2, a new genomic AI model. It predicts how DNA changes affect specific human cell types and states, helping researchers find disease-related genetic variants at the cellular level.

The model uses a new conditional architecture that takes tissue, cell type, or cell state as input. It was trained on over 107 million single-cell atlases from human and mouse samples, covering 68 tissues.

In tests on 28 types of primary immune cells, the model identified 30% of disease-causing variants, up from 16% before. Its accuracy in predicting gene expression changes rose from 23% to 43%.

The model is now open to non-commercial research through its API, while commercial users can apply for private preview. Google DeepMind notes it is for scientific research only and has not been validated for clinical or diagnostic use.