Google DeepMind has launched AlphaGenome 2, a new artificial intelligence model that predicts how changes in DNA affect gene activity in specific human cell types. The company announced the model on 11 November, and it is available now for non-commercial research through the AlphaGenome API.
The model takes a DNA sequence of up to one million letters and predicts how a variant in that sequence changes what a cell does.
Earlier models worked from broad tissue averages, so effects that appear in only one cell type were easy to miss. AlphaGenome 2 narrows the prediction to a single tissue, cell type or cell state at a time.
How the new model works
AlphaGenome 2 is built as a conditional model. It reads the DNA sequence together with a chosen biological condition, such as a gut macrophage or an activated immune cell. The same stretch of DNA can then be read differently for different cells.
The model can also work from a scientist’s own single-cell data. A researcher can upload gene activity from as few as a few dozen cells, and the model adapts to that unseen cell state without retraining. It can also predict measurements the scientist did not take, such as chromatin accessibility.
Nearly twice the variant recall
DeepMind tested the model on TenK10K, a large single-cell map of 28 primary immune cell types. AlphaGenome 2 identified 30 percent of the causal variants, against 16 percent for the first AlphaGenome model tuned on single-cell data.
It also called the direction of gene effects correctly in 43 percent of cases, against 23 percent before. That gap is the core of the upgrade, since knowing whether a variant turns a gene up or down is a key clue to how it drives disease.
The model was trained on more than 107 million individual cell profiles across 68 tissues, using human and mouse data.
DeepMind describes the result as a step-change improvement in variant prediction. In its announcement, the lab writes that AlphaGenome 2 “identifies nearly twice as many variants that change gene expression in cells, compared to our previous state-of-the-art model.”
Two real tests, from bowel disease to vaccines
The team checked the model against a known result in inflammatory bowel disease. A landmark study had shown that a variant in a distant regulatory switch drives the disease by overactivating the ETS2 gene in gut immune cells.
AlphaGenome 2 reproduced that finding on its own. It singled out the causal variant while leaving other cell types largely unaffected.
A second test used single-cell data from a COVID-19 mRNA vaccine study. The model told vaccine-activated immune T cells apart from resting ones, and it found that a variant linked to severe COVID-19 disrupts gene activity more strongly in the activated state.
Who can use it, and what it cannot do
AlphaGenome 2 is available now for non-commercial research through the AlphaGenome API, with commercial research in private preview.
DeepMind has also published precomputed predictions for trait-linked variants across hundreds of cell and tissue conditions. That covers traits from height to type 2 diabetes, and it lets researchers use the results without running large computations themselves.
The model is for research use only. DeepMind states clearly that neither AlphaGenome model has been validated or approved for clinical or diagnostic use, so the tool is a research instrument, not a medical one.
Where it fits in DeepMind’s science push
AlphaGenome 2 follows a fast-moving line of science work at DeepMind. The first AlphaGenome launched last year, and the AlphaGenome Atlas followed with predictions for all 9 billion possible single-letter changes in the human genome.
The lab is also building AlphaGenome 2 Skills for Google Antigravity, its agentic development platform, to pair the model with Gemini 4 as a science assistant.
The work leans on long-running collaborations, including with researchers at the Wellcome Sanger Institute, to map more cell types, developmental lineages and perturbed cell states.
DeepMind is also inviting the research community to share AlphaGenome 2 use cases by 30 November, with selected projects eligible for extra API quota and credits.
For KeeVurds readers, the timing is part of a wider Google week. DeepMind’s parent company shipped Gemini 4 Carbon only yesterday, and AlphaGenome 2 shows the same research engine turning toward biology. The question now is how quickly working scientists turn the new predictions into real discoveries.
