AlphaGenome 2 predicts how DNA changes affect specific human cells

According to Google DeepMind:

AlphaGenome 2 is a next-generation DNA sequence model designed to bridge the gap between single-cell biology and human genetics. Unlike the original version which analyzed broad tissues in parallel, AlphaGenome 2 uses a conditional architecture to evaluate how a million-letter DNA sequence behaves in one specific cell type or state at a time. Trained on over 107 million individual cell profiles from humans and mice, the model can predict gene expression, chromatin accessibility, and DNA methylation across specialized cells. It can also generalize to unseen cell states, such as those triggered by diseases or vaccinations, without requiring retraining.

In benchmark tests, AlphaGenome 2 successfully identified nearly twice as many causal variants and directional gene effects compared to previous tools. It has already demonstrated the ability to pinpoint complex disease drivers, such as identifying a regulatory variant that overactivates the ETS2 gene in gut immune cells to cause inflammatory bowel disease. The tool is now available for non-commercial research via an API, with commercial access offered in private preview, and future integration planned for Google's agentic development platform.