DeepMind's AlphaGenome 2 predicts DNA effects separately for each cell type

reported by 2 outlets· 2 articles · confidence: medium · first seen 2026-11-11 15:00 UTC

What this means for you

Nothing to act on commercially: access is through the API for non-commercial research, and no price, licence or commercial route is given. There are also no benchmark results and no comparison with the first AlphaGenome, so the per-cell-type claim cannot be checked yet. If you work in genomics, the API is the thing to look at.

Google DeepMind released AlphaGenome 2 on 11 November, a DNA model that predicts regulatory activity — which stretches of DNA switch genes on or off — and the effects of genetic variants, separately for each cell type. It succeeds the original AlphaGenome. MarkTechPost describes it as a conditional sequence model: the cell type is an input that steers the prediction. It is reachable through the AlphaGenome API for non-commercial research only. The Verge says the aim is to show how DNA changes affect different cells, a problem it calls notoriously difficult; neither account publishes any evaluation results.

Key facts

  • ·AlphaGenome 2 is the successor to Google DeepMind's earlier AlphaGenome model. source
  • ·It is described as a conditional DNA sequence model that predicts regulatory activity and variant effects per cell type. source
  • ·It is available through the AlphaGenome API for non-commercial research. source
  • ·Both accounts of the release are dated 11 November 2026. source

What the sources say

  • The Verge AI — Frames the release around disease biology and the goal of linking DNA changes to individual cell types.
  • MarkTechPost — Brief release note supplying the technical description, access route and non-commercial restriction missing elsewhere.

Sources

The original reporting. Follow these — they did the work.

← the wire