Google DeepMind has released AlphaGenome 2, an AI model that predicts how changes in DNA affect gene activity in specific cell types. DeepMind's post is dated Nov. 11, 2026. It says the model is available now to researchers through the AlphaGenome API.
What happened
DeepMind describes AlphaGenome 2 as a model that helps explain how cells read DNA, to better understand disease. It follows the original AlphaGenome and the AlphaGenome Atlas, which DeepMind describes as predictions for all 9 billion single-letter changes in DNA. See also DeepMind's AlphaProof Code work on Signal's protocol for another recent DeepMind research release.
What it can do
The same DNA change can behave differently depending on the cell. AlphaGenome 2 is a conditional model, so it predicts for one tissue, cell type or cell state at a time. DeepMind says it takes up to 1 million letters of DNA as input and adds a new readout for DNA methylation, a chemical mark on DNA. It was trained on more than 100 million single-cell profiles from 68 tissues, in human and mouse.
DeepMind's headline test uses TenK10K, a dataset covering 28 primary immune cell types. On that data, DeepMind says AlphaGenome 2 identified causal variants at 30%, against 16% for the original AlphaGenome fine-tuned on single-cell data and other tools. It also reports 43% against 23% for directional gene effects, which DeepMind calls "nearly twice as many." Both are DeepMind's own comparisons.
DeepMind also gives two examples. It says the model reproduced the effect of a variant linked to inflammatory bowel disease (near the ETS2 gene) in gut macrophages. On COVID mRNA vaccine data, it says the model told activated T cells from resting ones, and that a variant tied to severe COVID disrupts gene activity more in the activated state.
Who can use it
This is a research tool, not a consumer product. DeepMind says AlphaGenome 2 is "available today for non-commercial research via the AlphaGenome API, and in private preview for commercial research." API documentation is at alphagenomedocs.com. DeepMind describes access only through the research API, not an app. DeepMind also mentions precomputed predictions for trait-associated variants across hundreds of cell and tissue conditions.
What is not known
DeepMind's post does not claim clinical use, and nothing here is a diagnosis or treatment tool. The figures above are DeepMind's own, and we have not seen independent testing. We do not know how the model performs on cell types or conditions outside those DeepMind reports. For another Google AI release this week, see Gemini 4 Argon in the Gemini app. More in AI and apps.
What we checked
We read DeepMind's blog post on the morning of Nov. 11, 2026 and listed the paper PDF it links as a source. We did not test the API. We have not independently reviewed the paper, so the claims above are DeepMind's as stated.
