Google DeepMind released AlphaGenome 2 on November 11 through the AlphaGenome API for noncommercial research, with a private preview for commercial research. The model takes a DNA sequence of up to 1,048,576 base pairs and a biological condition, such as a cell type or state. It predicts how a change in the sequence could affect gene activity in that setting.
Google DeepMind reports that AlphaGenome 2 identified 30% of causal variants in an immune-cell benchmark, compared with 16% for the original model fine-tuned on single-cell data. Researchers can also supply gene activity measurements from their own experiment to study a cell state the model has never seen. The release is for research use, and clinical or diagnostic use has yet to be validated.
