Search
Gene expression states vary widely across different human cells, a fact that has long impeded the mapping of disease genes. Google DeepMind has launched AlphaGenome 2, which employs a conditional architecture that takes tissue, cell type, or cellular state as input. The model was trained on over 107 million human and mouse single-cell atlases spanning 68 tissues. Its ability to identify relevant variants is nearly twice that of the previous generation, with substantial improvements across multiple benchmark metrics. The model demonstrates generalization capability and can analyze cell types and states on which it has not been trained. It incorporates multiple epigenomic predictions and supports cross-modal inference, and is currently open to non-commercial research use.
-
C114 Communication Network -
Communication Home
