DeepMind Launches AlphaGenome 2, Predicting DNA Variant Effects Cell by Cell

What happened

DeepMind announced AlphaGenome 2 in its official blog on Nov 11. The model reads one million DNA letters at a time and, given one biological condition — a tissue, cell type or cell state — predicts how a variant acts; the first AlphaGenome only predicted tissue-level averages, while version 2 narrows resolution to individual cell types. Training drew on bulk and single-cell data from human and mouse samples, more than 107 million individual cell profiles across 68 tissues.

The evaluation numbers come from TenK10K, a large single-cell map of 28 primary immune cell types: at 90% precision, AlphaGenome 2 recovered 30% of putatively causal expression-linked variants, against 16% for the first model fine-tuned on single-cell data; it got the direction of effect right 43% of the time, against 23%. The model also reproduced an inflammatory bowel disease study, pinpointing the causal variant driving overactivation of the ETS2 gene in gut macrophages.

Non-commercial research can use the model today via the AlphaGenome API, while commercial research enters a private preview. DeepMind precomputed predictions for trait-associated variants across hundreds of cell and tissue conditions for researchers to use directly; AlphaGenome 2 Skills are coming to Google Antigravity. The model is for research use only and has not been validated for clinical or diagnostic use; the community use-case call closes on Nov 30.

Key facts

1. What it is: A next-generation DNA sequence model; one million DNA letters plus one biological condition per prediction, evaluated one condition at a time
2. Training data: More than 107 million individual cell profiles across 68 tissues, from human and mouse bulk and single-cell data
3. Benchmarks: On TenK10K, 30% of causal variants recovered vs 16%, and 43% correct direction vs 23%, at 90% precision
4. Disease check: Reproduced an inflammatory bowel disease finding, pinpointing the causal variant for ETS2 in gut macrophages
5. Availability: Open today to non-commercial research via the AlphaGenome API; commercial research in private preview
6. Limits: Research use only, not validated for clinical or diagnostic use; community use-case call closes Nov 30