✦
Summary
AlphaGenome 2 is Google DeepMind's next-generation DNA sequence model, announced in November 2026, that predicts how DNA variants affect gene activity in specific human cell types and states. It bridges single-cell biology and human genetics to pinpoint disease-relevant variants, identifying nearly twice as many causal variants as AlphaGenome 1. Researchers submit a DNA sequence of up to 1 million letters plus a biological condition or cell type, and receive predicted molecular readouts for that condition.
Key Features
Cell-type-specific variant effect prediction via a conditional model, one condition at a time
DNA sequence input of up to 1 million letters
Cell encoder that conditions on a scientist's own single-cell RNA-seq snapshot with no retraining
Cross-modality imputation, predicting unmeasured properties from one measurement
Predicts gene expression, histone modifications, chromatin accessibility and DNA methylation
Trained on more than 107M individual cell profiles across 68 human and mouse tissues
Precomputed variant scores for GWAS traits across hundreds of cell and tissue conditions
Generalises to unseen cell and disease states, such as vaccine-activated versus resting T cells
About AlphaGenome 2
The model is conditional: it predicts gene expression, histone modifications, chromatin accessibility and DNA methylation for the requested cell type, and its cell encoder can condition on a scientist's own single-cell RNA-seq snapshot of a few dozen cells with no retraining. It also performs cross-modality imputation, predicting unmeasured properties such as chromatin accessibility from a single measurement, and generalises to unseen cell and disease states, for example distinguishing vaccine-activated from resting T cells in a COVID-19 mRNA vaccine study. It was trained on more than 107 million individual cell profiles across 68 human and mouse tissues.
AlphaGenome 2 is available through the AlphaGenome API with an API key, the alphagenome Python client, and interactive Colab tutorials, with precomputed variant scores for GWAS traits across hundreds of cell and tissue conditions. The API is free of charge for non-commercial research, while commercial research runs through Google Cloud in private preview. It is a research model only and is not validated for clinical or diagnostic use.