Google DeepMind releases AlphaGenome 2, which predicts DNA variant effects in specific human cell types
The model finds nearly twice as many causal variants as the original on the TenK10K immune cell dataset (30% vs. 16%). It is for research use only, not clinical use.
What happened
Google DeepMind has released AlphaGenome 2, an AI model that predicts how small changes in DNA affect different types of human cells. Every cell carries the same DNA, but a heart cell and an immune cell switch on different genes. The new model looks at one cell type at a time instead of broad tissue averages.
Why it matters to you
Many diseases start with DNA changes that act only in certain cells. A tool that predicts those effects can help researchers find which changes matter and where.
The takeaway
Researchers can now test DNA variants against specific cell types, but the tool is not approved for diagnosing patients.
Evidence check
What the source says about the study behind this health claim.
- Tested on
- Lab or computer studyResults on real patients count for more than results on written or simulated cases.
- Peer reviewed
- No (preprint or press release)Peer review means outside experts checked the methods. Preprints and press releases have not had that check.
- Funded by
- Not statedA funder that sells the product has a stake in the result.
- Sample size
- 28 primary immune cell types in TenK10K; over 107 million training cell profilesResults from small studies are more likely to be chance.
What it means for you
The model is available through the AlphaGenome API with Colab tutorials for expression conditioning, and it can be conditioned on a user's own single-cell data via a cell encoder.
Commercial research access is in private preview through Google Cloud, and the model is not validated for clinical or diagnostic use.
Google DeepMind offers non-commercial access now and commercial access in private preview via Google Cloud; the post gives no pricing or revenue figures.