AlphaGenome 2: Deciphering how cells read DNA to better understand disease
The conditional model predicts how DNA sequence variations affect specific cell types and states, trained on over 107 million cell profiles across 68 tissues.
Key points
- AlphaGenome 2 shifts to a conditional architecture that takes a million-letter DNA sequence alongside specific cell types or experimental states.
- On the TenK10K dataset, it identified 30% of causal variants and 43% of directional gene effects, nearly double the original model.
- The model generalizes to new cell states without retraining when provided a single-cell RNA-seq snapshot from a few dozen cells.
- Predictions now include DNA methylation alongside gene expression, histone modifications, and chromatin accessibility.
- Access is open for non-commercial research via the AlphaGenome API and in private preview for commercial research.
Why it matters
Genetics and disease researchers can pinpoint the cell-specific regulatory effects of DNA variants without running exhaustive experimental screens.
Sources
- AlphaGenome 2: Deciphering how cells read DNA to better understand disease (Google DeepMind, Nov 11, 2026, 3:02 PM)