Google DeepMind announced AlphaGenome 2 on November 11, 2026, a DNA sequence model that predicts how genetic variants change activity in specific human cell types rather than in broad tissue averages. The post is bylined by Adam Gayoso, Natasha Latysheva, Pushmeet Kohli and Žiga Avsec. According to Google DeepMind, the model is available now for non-commercial research through the AlphaGenome API and in private preview for commercial research.
The first AlphaGenome, launched last year, read a one-million-letter DNA sequence and predicted properties across thousands of experiments from many tissues at the same time. Google DeepMind says that parallel design spread the model's capacity thin and made closely related cell types hard to tell apart. AlphaGenome 2 is built differently. It is a conditional model that receives a target biological condition, meaning a tissue, a cell type or a cell state, at every stage of its architecture and predicts one condition at a time. The condition can be picked from a reference catalog or generated from a scientist's own single-cell experiment through a cell encoder.
Training drew on bulk and single-cell data from human and mouse samples, including more than 107 million individual cell profiles across 68 tissues. The outputs cover the readouts of the first model, such as gene expression, histone modifications and chromatin accessibility, and add DNA methylation, which Google DeepMind describes as widely requested. Because the model separates a cell's state from its outputs, it can also fill in measurements that were never taken. A researcher who supplies only gene activity, for example, can get predicted chromatin accessibility back, which may help decide which experiment to run next.
The headline benchmark result comes from TenK10K, a single-cell map of 28 primary immune cell types. Google DeepMind reports that AlphaGenome 2 identified 30 percent of causal variants against 16 percent for the original AlphaGenome fine-tuned on single-cell data and other specialized tools, and recovered the direction of gene effects in 43 percent of cases against 23 percent. In one case study it reproduced a published finding that a distant regulatory variant drives inflammatory bowel disease by overactivating the ETS2 gene in gut macrophages, singling out the causal variant and showing that it boosts ETS2 in those cells while leaving other cell types and genes largely unaffected.
The company also says the model can adapt to new cell states without retraining. A scientist uploads single-cell RNA-seq data from as few as a few dozen cells, the activity is compressed into a cell-state representation, and the model is conditioned on it. In an example built on unseen data from a COVID-19 mRNA vaccine study, AlphaGenome 2 separated vaccine-activated T cells from resting ones and indicated that a variant linked to severe COVID-19 disrupts gene activity more strongly in the activated state, an effect that a resting-cell model would miss.
Google DeepMind has also precomputed predictions for trait-associated variants across hundreds of human cell and tissue conditions. The aim is to map enhancer-gene links in primary human cell types, such as brain astrocytes, where CRISPR screens on lab-grown cell lines rarely reach, and to narrow genome-wide association signals for traits ranging from height to type 2 diabetes down to specific variants and the cells where they act. Quoted in the post are Professor Sarah Teichmann of the University of Cambridge, a co-founder of the Human Cell Atlas consortium, and Peter Donnelly, CEO of Genomics Limited, who said his company looks forward to using it in its Mystra platform.
The caveats are stated by Google DeepMind itself. AlphaGenome and AlphaGenome 2 are intended for research use only and have not been validated or approved for clinical or diagnostic purposes. The accuracy figures come from the company's own announcement rather than from outside replication. Google DeepMind adds that AlphaGenome 2 Skills for its Antigravity agentic development platform are planned for release soon.