From the source
Google DeepMind announced AlphaGenome 2 on November 11, 2026, a deep-learning model that predicts how DNA variants affect gene regulation in individual cell types and states.
- The model is trained on more than 100 million individual cell profiles and is available for non-commercial research, with commercial research in private preview.
- It addresses a blind spot in tissue-wide genomic models, where a variant that matters only in a specialized or disease-activated cell has its signal diluted in a tissue average.
- AlphaGenome 2 conditions each prediction on a selected tissue, cell type or cell state, including conditions generated from a researcher's own single-cell experiment, and adds CpG DNA methylation to the original model's readouts.
- On the TenK10K single-cell immune dataset spanning 28 cell types, DeepMind's paper reports it recovered 30% of putatively causal expression-linked variants at 90% precision, against 16% for AlphaGenome 1 fine-tuned on single-cell data.
- DeepMind precomputed an enhancer-gene map across 640 primary single-cell reference conditions and scores for more than 6.8 million variants across 510 human conditions.
- The paper cautions that the model recovers only a minority of nominated variants, can underestimate effects outside accessible chromatin, and that condition representations capture correlations rather than causal relationships.