DeepMind's new genome model predicts DNA effects cell type by cell type

AlphaGenome 2 conditions its predictions on a chosen tissue or cell type, and DeepMind reports it beats its predecessor at spotting causal variants in immune cells.

Google DeepMind announced AlphaGenome 2 on November 11, 2026. It is a model that predicts how DNA sequences and variants affect specific human cell types. The original AlphaGenome, launched the previous year, predicted variant effects on molecular activity across broad tissues.

The new version is conditional. At every stage of its architecture it takes a 1-million-letter DNA sequence plus one target condition, such as a tissue, cell type or cell state. That condition can come from a reference catalog or be generated from a user's own single-cell experiment. The model was trained on bulk and single-cell data from human and mouse, including over 107 million individual cell profiles across 68 tissues. It predicts gene expression, histone modifications, chromatin accessibility and, newly, DNA methylation.

On the TenK10K dataset of 28 primary immune cell types, DeepMind reports it identified causal variants at 30% versus 16%, and directional gene effects at 43% versus 23%, compared with the original model fine-tuned on single-cell data and other tools. It also reproduced a published finding that a distant regulatory variant drives inflammatory bowel disease by boosting ETS2 in gut macrophages. Given a few dozen cells of single-cell RNA data, it can condition on unseen cell states without retraining. With COVID-19 mRNA vaccine data, it found that a severe-COVID-linked variant disrupts gene activity more in activated T cells than in resting ones.

Why it matters: many disease-linked variants sit in regions where the effect only shows in particular cell types. Predicting that computationally, before running experiments, could cut the cost and time of finding disease mechanisms, and DeepMind has precomputed predictions for trait-associated variants across hundreds of cell and tissue conditions, from height to type 2 diabetes.

Caveats: the performance figures are DeepMind's own. The tool is for research only and is not validated or approved for clinical or diagnostic use. Access is through the AlphaGenome API for non-commercial research now, with a private preview for commercial research. Work with the Wellcome Sanger Institute to cover more cell types is ongoing.

Source: Google DeepMind