DeepMind Releases AlphaGenome 2 for Cell-Type DNA

Google DeepMind on 11 November 2026 released AlphaGenome 2, a DNA-sequence model that predicts how variants affect specific human cell types rather than broad tissue averages. In a DeepMind blog post by Adam Gayoso, Natasha Latysheva, Pushmeet Kohli and Žiga Avsec, the lab said the model “predicts DNA effects for one condition — defined as a tissue, cell type or state — at a time,” and was trained on “over 107 million individual cell profiles.” A summary line in the same post also says “more than 100 million … across 68 tissues.” It is available “for non-commercial research via the AlphaGenome API, and in private preview for commercial research.”

The same day, at 13:49 GMT (Europe/London), the google-deepmind/alphagenome GitHub repository tagged v2.0.0. The release notes add an AlphaGenome 2 model API (alphagenome.models.v2) “supporting reference-condition and single-cell expression-conditioned predictions, variant and interval scoring, cell embeddings, and input gradients,” plus tutorials for the new model.

DeepMind is explicit about the limit that matters most for readers: “AlphaGenome and AlphaGenome 2 are intended for research use only and have not been validated or approved for clinical or diagnostic purposes.” There is no peer-reviewed paper linked from the announcement. The performance figures below are DeepMind’s own tests on past datasets.

A conditional model, and DeepMind’s own scores

The first AlphaGenome took a million-letter DNA sequence and predicted properties across many tissues at once. AlphaGenome 2 is a conditional model: the target biological condition is an input at every stage, alongside that sequence. The condition can be chosen from a reference catalogue of cell types or generated from a scientist’s own single-cell experiment through a cell encoder. DeepMind says this also allows “cross-modality imputation,” so a scientist who inputs one measurement, such as single-cell gene activity, can obtain predictions for unmeasured properties such as chromatin accessibility. The model still predicts gene expression, histone modifications and chromatin accessibility, and it adds DNA methylation, “a widely requested readout.” Training used bulk and single-cell data from human and mouse samples.

On the TenK10K dataset, a single-cell map of 28 primary immune cell types, DeepMind says AlphaGenome 2 “reliably identified nearly twice as many causal variants (30% vs. 16%) and directional gene effects (43% vs. 23%) as the original AlphaGenome fine-tuned on single-cell data and other specialised tools.” Those percentages are the lab’s own comparison. It also says the model computationally reproduced a known inflammatory-bowel-disease finding: a distant regulatory variant that overactivates the ETS2 gene in gut macrophages, “leaving other cell types and genes largely unaffected.”

A further claim is that the model can adapt to new single-cell data “from just a few dozen cells” without retraining. Professor Sarah Teichmann, co-founder of the Human Cell Atlas and professor of stem cell medicine at the University of Cambridge, is quoted: “What makes AlphaGenome 2 exciting is how it adapts directly to new single-cell data without retraining, predicting how DNA variants act in newly profiled cell states from even a small sample of cells.” DeepMind says that, given unseen data from a COVID-19 mRNA vaccine study, the model distinguished vaccine-activated T cells from resting ones and found that a variant linked to severe COVID-19 disrupted gene activity more strongly in the activated state. Peter Donnelly, chief executive of Genomics Limited, said DNA sequence prediction models help move from a GWAS association “to what a drug programme needs,” and that the company looks forward to “incorporating it in our Mystra platform.” That is a planned use, not a completed clinical integration, and it does not change the research-only limit.

DeepMind has also precomputed predictions for trait-associated variants “across hundreds of human cell and tissue conditions,” including enhancer–gene mapping in primary cells such as brain astrocytes. AlphaGenome 2 Skills in Google Antigravity are promised “soon.”

Better variant recall is not a diagnostic test

The scientific step here is real enough to report carefully: moving from tissue averages to a model that conditions on a named cell type, and that can be pointed at a small new single-cell snapshot. The TenK10K lifts, the ETS2 reproduction and the vaccine-state example are all DeepMind’s own accounts of how that architecture behaves on data it has already seen or can be shown. They are not an independent clinical validation, and DeepMind does not claim they are. The sentence in the blog that the models “have not been validated or approved for clinical or diagnostic purposes” is the one that should travel with every mention of disease-relevant variants.

That is the same caution TSN has had to keep next to DeepMind’s other biology tools. The Antigravity agent’s antimicrobial-peptide candidate was a finding that still needed lab testing. Cheminformatics skills added to Antigravity are another research workflow, not a regulated assay; AlphaGenome 2 Skills, when they appear, will sit in that same agentic-science track. DeepMind’s programme giving up to 200 European researchers access to frontier models is the access route some academic users may already be on, including in the UK, where Teichmann’s Cambridge affiliation and the Wellcome Sanger Institute collaboration named in the post give the announcement a direct research-community hook. None of those programmes converts a research API into a diagnostic device.

The facts to watch are a peer-reviewed paper, the opening of the commercial private preview to named users, and the Antigravity skills release. Until then, AlphaGenome 2 is a cell-type-specific DNA model, available for non-commercial research through the official API, and it remains a research tool, not a clinical test.

Sources

Adam Gayoso, Natasha Latysheva, Pushmeet Kohli and Žiga Avsec, Google DeepMind, “AlphaGenome 2: Deciphering how cells read DNA to better understand disease,” 11 November 2026. https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/
google-deepmind/alphagenome, GitHub release v2.0.0, 11 November 2026, 13:49 GMT (Europe/London). https://github.com/google-deepmind/alphagenome/releases/tag/v2.0.0