DeepMind’s AlphaGenome 2 predicts DNA effects by cell type

The same DNA variant can affect different cells in different ways, depending on which genes are active and what a cell is doing. Google DeepMind’s new model, AlphaGenome 2, is designed to predict those effects at the level of individual cell types and states.

Corp and Tech·November 11, 2026·reads: 779

DeepMind’s AlphaGenome 2 predicts DNA effects by cell type

The model succeeds AlphaGenome, which DeepMind released last year. Although nearly every cell in the body carries the same DNA, heart muscle, skin and immune cells use different genes. Those patterns can also shift with a cell’s state, making it difficult to assess a genetic variant’s effects across the body.

Google said in a blog post published Wednesday that AlphaGenome 2 offers a finer-grained view than broader models such as AlphaGenome. That detail could help researchers identify disease-related effects in specific cells that might otherwise be missed.

DeepMind says the model was trained on more than 100 million individual cell profiles across 68 tissues. The company describes its approach as turning static cell atlases — repositories that map and catalog cells — “into a predictive engine for human genetics.”