AI is learning to read DNA one cell type at a time. On November 11, Google DeepMind announced AlphaGenome 2, a new model that predicts how a change in your DNA affects specific cells — not just whole tissues averaged together.
What changed
- Cell by cell, not tissue averages: the first AlphaGenome read DNA across tissues all at once. AlphaGenome 2 works on one cell type or cell state at a time, so a change that only matters in one kind of cell no longer gets washed out.
- Trained on a lot of cells: DeepMind says it learned from more than 107 million individual cell profiles across 68 tissues, from human and mouse samples.
- Nearly twice as many finds: on a test set of 28 immune cell types, it spotted 30% of the DNA variants believed to change gene activity, up from 16% for the earlier model — and it called the direction of the gene effect 43% of the time, up from 23%.
- Your own data counts: researchers can pick a cell type from a catalog, or feed in their own single-cell experiment and have the model adapt to it.
Try it like this
- If you do research, it is available now for non-commercial work through the AlphaGenome API — commercial research is in private preview.
- Start with one clear question: pick one DNA change and one cell type, and ask what the model predicts there.
- Compare cell types. The same change can do nothing in one cell and matter a lot in another — that difference is the whole point of this version.
- Treat every result as a lead, not an answer. A prediction tells you where to look in the lab next.
Good to know
- Research only: DeepMind is clear that AlphaGenome 2 has not been validated or approved for clinical or diagnostic use. It cannot diagnose anyone.
- Still early: 30% is a big jump, but it also means most likely variants are still missed. This is a tool for narrowing the search, not settling it.
The tip: when you read AI health headlines, check which cell type the claim is about — with this generation of models, the cell is where the real story is.
