Google DeepMind releases AlphaGenome 2, predicting DNA variant effects in specific cell types
Google DeepMind released AlphaGenome 2 on 11 November 2026. It predicts how DNA changes act in specific cell types and is free for non-commercial research.
By Emre A.Published
Tech AI LatestGoogle DeepMind released AlphaGenome 2 on 11 November 2026, a DNA sequence model that predicts how genetic variants change gene regulation in individual human cell types rather than in whole tissues. Trained on more than 100 million single-cell profiles from human and mouse samples, it found 30% of causal variants on the TenK10K immune-cell benchmark, against 16% for its predecessor, DeepMind says. Non-commercial researchers can use it free through the AlphaGenome API.
Key facts
- Announced: 11 November 2026 (Google DeepMind blog)
- Training data: bulk and single-cell data from human and mouse, more than 100 million individual cell profiles
- TenK10K benchmark (28 immune cell types): 30% of causal variants found at 90% precision, against 16% for AlphaGenome 1 fine-tuned on single-cell data
- Access: free for non-commercial research through the AlphaGenome API; commercial research by application to a private preview
- Research use only; not validated or approved for clinical or diagnostic use, DeepMind says
- Not stated: commercial pricing, API quotas, release date for the planned Antigravity skills
What Google DeepMind released
Google DeepMind released AlphaGenome 2 on 11 November 2026, the second generation of its DNA sequence model. The announcement is signed by Adam Gayoso, Natasha Latysheva, Pushmeet Kohli and Žiga Avsec. Where the first AlphaGenome predicted how a change in the DNA code alters gene regulation across tissues, the new model makes those predictions for individual human cell types, which is the level at which many diseases actually start.
DeepMind says AlphaGenome 2 was trained on bulk and single-cell data from human and mouse samples, covering more than 100 million individual cell profiles. The blog post describes a model that bridges single-cell biology and human genetics, predicts variant effects across hundreds of human cell types, adds DNA methylation to the outputs the first version produced, and can adapt to new single-cell datasets without being retrained.
The authors explain the motivation directly: the first AlphaGenome spread its capacity across many tissues, which made it hard to capture the subtle differences between closely related cell types, for example two kinds of immune cell that live side by side in the blood but switch genes on and off differently.
The benchmark numbers
The headline result comes from TenK10K, a dataset that links genetic variants to gene expression in 28 primary immune cell types. At a precision of 90%, AlphaGenome 2 recovered 30% of causal variants, the DNA changes that actually drive a difference in gene activity. AlphaGenome 1 fine-tuned on single-cell data recovered 16% under the same threshold, and DeepMind says other specialised tools did no better than that. In plain terms, the new model finds close to twice as many real regulatory variants for the same rate of false alarms.
These are the developer's own figures on a benchmark it selected, and they measure prediction quality, not patient outcomes. Independent replication by outside genetics groups had not been published at the time of writing. The 30% figure is also a recall number: most causal variants in the set are still missed, which matters for anyone planning to use the model to rank candidates in a study.
Who can use it, and on what terms
Non-commercial researchers can call AlphaGenome 2 now through the existing AlphaGenome API, free of charge. Companies doing commercial research have to apply for a private preview. The model weights are not released, so all use goes through Google's servers. DeepMind states that AlphaGenome 2 is intended for research only and has not been validated or approved for clinical or diagnostic purposes; it is not designed to diagnose disease.
DeepMind also says it will soon release AlphaGenome 2 skills for Google Antigravity, its agentic development platform, so that coding and research agents can call the model as one step in a larger analysis. No date was given. Pricing for commercial access and API quotas were not stated in the announcement.
How this fits DeepMind's genomics work
The original AlphaGenome was previewed in June 2025 and published in Nature on 28 January 2026. It reads up to one million DNA letters at a time and predicts thousands of functional signals, and the paper reported that it matched or beat the best existing models in 25 of 26 variant effect evaluations. In September 2026 DeepMind used that model to build AlphaGenome Atlas, a 1-petabyte set of precomputed scores for all 9 billion possible single-letter changes in the human genome.
AlphaGenome 2 moves the question from whether a variant matters to where it matters. That is the step most relevant, among the AI research results of recent months, to drug target discovery and to interpreting genome-wide association studies, where a signal near a gene often has to be traced to the one cell type in which it acts. The open questions are how the model performs outside immune cells, whether outside labs reproduce the TenK10K gain, and what commercial users will pay once the private preview ends.
What changed
| Before | AlphaGenome 1 predicted variant effects at the level of tissues and recovered 16% of causal variants on TenK10K when fine-tuned on single-cell data |
|---|---|
| After | AlphaGenome 2 predicts effects in individual cell types, adds DNA methylation, and recovers 30% of causal variants on TenK10K at 90% precision, DeepMind says |
| Who it affects | Academic geneticists (free API access), drug discovery teams (private preview), developers using Google Antigravity agents |
Questions people ask
- What is AlphaGenome 2?
- A DNA sequence model from Google DeepMind, released on 11 November 2026, that predicts how genetic variants change gene regulation in specific human cell types. It was trained on more than 100 million single-cell profiles.
- Is AlphaGenome 2 free to use?
- Yes for non-commercial research, through the AlphaGenome API. Commercial researchers must apply for a private preview, and prices for that have not been published.
- Can AlphaGenome 2 be used to diagnose patients?
- No. Google DeepMind says the model is for research use only and has not been validated or approved for clinical or diagnostic purposes.
Sources
- AlphaGenome 2: Deciphering how cells read DNA to better understand disease (Google DeepMind, 11 Nov 2026)
- Google DeepMind Releases AlphaGenome 2 To Study Disease (Dataconomy, 12 Nov 2026)
- AlphaGenome 2 finds nearly 2x more disease-linked DNA variants (Interesting Engineering, 11 Nov 2026)
- Advancing regulatory variant effect prediction with AlphaGenome (Nature, 28 Jan 2026)
Reference pages
- AlphaGenome 2 model
- Google DeepMind company
Spotted an error? See the corrections policy and tell the editor. Facts in this story come from the sources listed above; vendor claims are attributed.
Comments
Loading comments.





