THE CRUNCH
Google DeepMind has unveiled AlphaGenome 2, an AI model that predicts how changes in DNA affect individual cell types and states rather than treating the genome broadly. Because nearly every cell carries the same DNA but switches different genes on and off, the same genetic variant can have very different effects depending on the cell. The model is designed to capture those differences, which DeepMind says could reveal disease-relevant effects in specific cells that broader models might miss.
AlphaGenome 2 was trained on more than 100 million individual cell profiles across 68 tissues, effectively turning static cell atlases, the large repositories that catalogue individual cells, into what DeepMind calls a predictive engine for human genetics. The company says it identifies nearly twice as many variants that can change gene expression in cells as its previous state-of-the-art model.
The researchers also report that the model can make predictions about new cell types and biological states without retraining, and can spot gaps in experimental data, potentially helping scientists decide what to investigate next. DeepMind's genomics lead Žiga Avsec told The Verge that this granularity matters for drug development, since drugs target specific proteins and specific cell types.
There is a trade-off: AlphaGenome 2's finer-grained predictions are more computationally expensive than its predecessor's, so it is not replacing AlphaGenome. Those demands would have made it impractical to produce the genome-wide Atlas DeepMind released in September, which covered every possible DNA letter change in the human genome.
AlphaGenome 2 is available for non-commercial research via the AlphaGenome API from Wednesday, with a limited private preview for commercial research.