Overview
- DeepMind announced AlphaGenome 2 on November 11, 2026 as a successor to AlphaGenome that predicts how non-coding DNA variants change gene activity in specific cell types and states.
- The model was trained on more than 100 million single-cell profiles from 68 tissues in human and mouse, letting it read the same DNA differently depending on a target cell condition.
- DeepMind reports that AlphaGenome 2 finds nearly twice as many expression-altering variants in single-cell tests as the prior model, citing roughly 30% recall versus 16% before, but those numbers come from company benchmarks focused on blood and immune cells.
- The team used AlphaGenome 2 to rapidly reproduce aspects of a 2024 lab result linking an ETS2 enhancer variant to inflammatory bowel disease, showing the model can generate testable hypotheses far faster than wet-lab work.
- Access is available for non-commercial research through the AlphaGenome API with a limited commercial preview, yet higher computational cost limits some large-scale uses and independent experimental validation is needed to confirm broad biological utility.