Google DeepMind launches AlphaGenome 2 to predict how DNA changes affect specific cell types
Trained on more than 100 million single-cell profiles across 68 tissues, the model finds nearly twice as many expression-changing variants as its predecessor and is available for non-commercial research.
Why it matters
The same genetic variant can have very different effects depending on cell type and state, which is why most disease-linked variants are hard to interpret. AlphaGenome 2 moves from broad tissue-level predictions to individual cell types and states, and DeepMind says it can make predictions for new cell types without retraining and point to gaps in experimental data. It does not replace the first AlphaGenome, because its finer predictions cost more compute. Watch for independent validation, uptake by drug developers and the multi-species and virtual-cell work DeepMind describes as next.
Key facts
- Trained on 100M+ single-cell profiles
- Covers 68 tissues
- Finds nearly 2x as many expression-changing variants as predecessor
- Predicts new cell types without retraining
- Available for non-commercial research
- Successor to AlphaGenome, released last year