Skip to content

Research Story

AlphaGenome 2: DeepMind unveils a cell-specific model to predict how DNA variants alter gene activity

Google DeepMind announced AlphaGenome 2 on 11 November 2026, a conditional DNA sequence model that predicts variant effects in specific cell types and states, which the company says was trained on over 107 million single-cell profiles and is not validated for clinical use.

Published

This story was written with the help of artificial intelligence, checked automatically, then reviewed by an editor before publication, under our AI policy.

What DeepMind announced

Google DeepMind announced AlphaGenome 2 in a blog post dated 11 November 2026. It is a DNA sequence model that predicts how a sequence and its variants influence specific cell types and cellular states. ZDNET France reported the announcement on 12 November 2026; the details below come from DeepMind's own post unless stated otherwise.

The model is conditional. It takes a biological condition (a tissue, cell type or cell state) together with a DNA sequence, and predicts effects for one condition at a time. DeepMind contrasts this with the previous model, which made predictions for several conditions in parallel.

AlphaGenome 2 predicts the same experimental readouts as AlphaGenome (gene expression, histone modifications and chromatin accessibility) and adds DNA methylation.

Training data and claimed performance

According to DeepMind, the model was trained on more than 100 million individual cell profiles covering 68 tissues. The post cites "over 107 million individual cell profiles" from human and mouse samples.

The performance figures below are claims by the company, not independent results. DeepMind says AlphaGenome 2 "identifies nearly twice as many variants that change gene expression in cells" as its previous state-of-the-art model.

On the TenK10K dataset, which covers 28 immune cell types, DeepMind reports that AlphaGenome 2 recovers causal variants in 30% of cases. The comparison figure is 16%, for the original AlphaGenome fine-tuned on single-cell data and for other specialised tools. The post does not detail, in the material available here, how these benchmarks were run or whether they have been replicated outside DeepMind.

Examples given by DeepMind

DeepMind says AlphaGenome 2 computationally reproduced a study on inflammatory bowel disease, in which a variant causes the gene ETS2 to be overactive in intestinal macrophages.

The company also says the model can adapt to new single-cell data without retraining, using a small number of cells. On data from a Covid-19 mRNA vaccine, it reportedly distinguished activated T cells from resting ones.

Access and limits

AlphaGenome 2 is available for non-commercial research through the AlphaGenome API, and in private preview for commercial research. ZDNET France adds that non-commercial access is free. DeepMind also announces AlphaGenome 2 "Skills" for Google Antigravity, to come; no date is given in the facts available.

DeepMind states that AlphaGenome and AlphaGenome 2 "have not been validated or approved for clinical or diagnostic purposes". The tool is therefore a research instrument, and the results above do not establish clinical usefulness.

TenK10K benchmark: causal variants recovered (DeepMind's reported figures) (data as of )
ModelCausal variants recovered
AlphaGenome 230%
Original AlphaGenome fine-tuned on single-cell data and other specialised tools16%

FAQ

What does AlphaGenome 2 predict?

It predicts how a DNA sequence and its variants affect specific cell types and states, including gene expression, histone modifications, chromatin accessibility and DNA methylation.

Can AlphaGenome 2 be used for diagnosis?

No. DeepMind says the model has not been validated or approved for clinical or diagnostic purposes.

Who can access it?

Non-commercial researchers can use it via the AlphaGenome API; commercial researchers can join a private preview.

Sources

  1. https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/, deepmind.google, (primary source)
  2. Google DeepMind dévoile AlphaGenome 2 : une IA pour comprendre comment les cellules lisent l’ADN, ZDNET France,

Plain text version (Markdown)