DeepMind Unveils AlphaGenome 2 Doubling Pathogenic Variant Detection

Google DeepMind has released AlphaGenome 2, a next-generation DNA sequence model that refines variant effect prediction from tissue averages to individual cell types and states, nearly doubling the number of detectable gene-regulatory variants compared to previous best models. This breakthrough enables scientists to systematically address why the same variant can have divergent effects across different cells, offering a new tool to pinpoint disease drivers at the cellular level.

The model employs a conditional architecture trained on over 107 million single-cell profiles, capturing subtle differences between closely related cells and supporting cross-modal data imputation to predict unmeasured properties. DeepMind has made the model available via API for non-commercial research and plans to integrate it into the Google Antigravity agent platform.

The release marks a key step in AI-driven genomic regulation analysis, potentially accelerating translational research from genetic variants to disease mechanisms and advancing precision medicine and therapeutic discovery. Its single-cell resolution provides new insights into cellular heterogeneity in complex diseases, with far-reaching implications for genetics, cell biology, and clinical research.