Google DeepMind releases AlphaGenome 2, capable of predicting the impact of genetic variants by cell type.

Beating AI News Flash: Google DeepMind announces the launch of AlphaGenome 2, a next-generation DNA sequence model designed to predict how DNA sequences and their variants affect specific human cell types, helping to study the cellular-level causes of disease. Building on last year's AlphaGenome, the model refines predictions from coarser tissue averages to individual cell types and cell states, and features a redesigned model architecture. According to the official announcement, compared with the previous state-of-the-art model, AlphaGenome 2 identifies nearly twice as many variants that can alter gene expression in cells. The model was trained on human and mouse population and single-cell data, including over 107 million single-cell atlases covering 68 tissues, and can transform static cell atlases into a prediction engine for human genetics.

AlphaGenome 2 is a conditional model that makes predictions for one tissue, cell type, or cell state at a time, and can also adapt to new single-cell data provided by researchers without retraining, and infer missing experimental data. The model is available starting today through the AlphaGenome API for non-commercial research, while commercial research is offered in a limited, non-public preview. The official statement emphasizes that AlphaGenome and AlphaGenome 2 are for research use only and have not been approved for clinical or diagnostic purposes.