Google DeepMind announced AlphaGenome 2 on 11 November, a DNA model that predicts how genetic variants affect gene regulation in individual cell types and states โ€” not tissue averages. Where earlier models blurred the effects of a variant across a whole tissue sample, AlphaGenome 2 aims to show what a variant does in each specific cell, a shift that matters because most disease-linked variants act in only one or a few cell types.

The model was trained on more than 107 million individual cell profiles across 68 tissues from humans and mice. On the TenK10K immune-cell benchmark, it recovered 30% of putatively causal expression-linked variants at 90% precision, compared with 16% for the original AlphaGenome โ€” nearly double the hit rate at the same precision.

AlphaGenome 2 is available from today for non-commercial research through the AlphaGenome API, while commercial research access is in private preview. DeepMind positions the model as a research tool for understanding how cells read DNA and how that reading goes wrong in disease.

One important limit: AlphaGenome 2 is for research use only. It is not validated or approved for clinical or diagnostic use, so its predictions are a starting point for scientists, not a medical answer.

Sources