Google DeepMind has announced AlphaGenome 2, an artificial intelligence model that predicts how DNA sequences and their variations affect specific human cell types. Every cell in the body carries the same DNA blueprint, yet each cell type - such as a heart muscle cell or an immune cell - switches different genes on or off to do its job. Predicting how small DNA variations change this gene activity inside individual cells has been an unsolved challenge, and such variations are a key driver of human traits and disease.
The model succeeds the original AlphaGenome, released last year, which predicted the molecular effects of DNA variants only as broad tissue averages. AlphaGenome 2 is a conditional model: it takes a chosen tissue, cell type or cell state as an extra input alongside a million-letter DNA sequence, and predicts how that particular condition reads the DNA. It was trained on more than 107 million individual cell profiles from human and mouse samples across 68 tissues, and it adds DNA methylation, a chemical mark that helps control gene activity, to its predictions.
Sharper predictions for disease research
On TenK10K, a large single-cell map of 28 immune cell types, AlphaGenome 2 identified nearly twice as many causal variants (30 percent versus 16 percent) and directional gene effects (43 percent versus 23 percent) as the previous model. It also reproduced a known finding about a variant that boosts the ETS2 gene in gut immune cells called macrophages, while leaving other cell types largely unaffected. In another test, the model adapted to unseen data from a COVID-19 mRNA vaccine study without retraining, and showed that a variant linked to severe COVID-19 disrupts gene activity more strongly in vaccine-activated immune cells than in resting ones.
This matters because genome-wide association studies, scans that link stretches of DNA to a disease, often cannot identify the exact causal variant, its target gene or the cell type where it acts. DeepMind has precomputed AlphaGenome 2 predictions for trait-associated variants across hundreds of cell and tissue conditions, covering traits from height to type 2 diabetes. The model is available for non-commercial research through the AlphaGenome API, and in private preview for commercial research. It is intended for research use only and has not been validated or approved for clinical or diagnostic purposes.