AlphaGenome 2 Maps DNA Effects Cell-by-Cell

AlphaGenome 2 makes genomic AI substantially more useful by shifting from broad tissue-level averages to cell-specific variant hypotheses, though its predictions remain research aids—not clinical conclusions.

• –Its conditional architecture accepts a million-letter DNA sequence plus a target tissue, cell type, or cellular state.
• –On the TenK10K immune-cell benchmark, DeepMind reports 30% of causal variants identified versus 16% by the previous model, with directional gene effects at 43% versus 23%.
• –Researchers can condition predictions on their own single-cell data and impute unobserved biological modalities, opening practical workflows for variant prioritization and hypothesis generation.
• –The API and documentation expose programmatic variant scoring, sequence prediction, cell encoding, and visualization workflows.
• –Research-only access, variable query rates, and the absence of clinical validation limit immediate production or diagnostic use. [AlphaGenome documentation](https://www.alphagenomedocs.com/)