Skip to content

Research Story

AlphaGenome 2: DeepMind unveils a model predicting how DNA variants act in specific human cell types

Google DeepMind announced AlphaGenome 2 on 11 November 2026, a DNA sequence model that predicts how genetic variants affect specific human cell types and states, trained on more than 100 million cell profiles across 68 tissues.

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 has announced AlphaGenome 2, a DNA sequence model that predicts how genetic variants affect specific human cell types and cell states (DeepMind announcement, 11 November 2026).

The model is conditional: it predicts the effects of DNA for one condition at a time (a tissue, a cell type or a cell state), rather than predicting in parallel. According to DeepMind, it can also process new single-cell data without retraining and predict variant effects for cell states it has not seen before.

Nature reports that AlphaGenome 2 was trained on measurements from more than 100 million individual cells across 68 tissues, and that it predicts across 637 human and 311 mouse cell types. DeepMind's own wording is that it was trained on more than 100 million individual cell profiles across 68 tissues.

Claimed performance

According to DeepMind, AlphaGenome 2 identifies nearly twice as many variants that change gene expression in cells as its previous state-of-the-art model. These are company claims; the facts available do not include an independent evaluation.

One benchmark figure is given. On the TenK10K dataset, which covers 28 immune cell types, DeepMind reports that the model identified 30% of causal variants, against 16% for the original AlphaGenome fine-tuned on single-cell data.

DeepMind also says the model computationally reproduces a 2024 finding: a variant in a distant regulatory switch that overactivates the gene ETS2 in gut macrophages, which is implicated in inflammatory bowel disease (IBD).

Reaction from researchers

Nature reports that the team of James Lee at the Francis Crick Institute, who made the IBD/ETS2 discovery, said the model replicated most of it in 30 minutes, compared with a year and a half of experiments. Nature quotes the comment: “They've managed to figure out the answer in 30 minutes”.

Nature also reports that the earlier AlphaGenome 1 did not specify the cell type in which a variant acts. Lee considers this a problem for its usefulness.

According to The Verge, Žiga Avsec said that resolution by cell type matters for targeting specific proteins and cell types in drug development.

Availability and limits

AlphaGenome 2 is available for non-commercial research through the AlphaGenome API, and in preliminary private access for commercial research.

DeepMind states that AlphaGenome and AlphaGenome 2 are for research only and have not been validated or approved for clinical or diagnostic use.

The Verge describes AlphaGenome 2 as the successor to AlphaGenome, released the previous year. It reports that the new model will not replace the original in the short term because its predictions are more costly in computation. The facts available do not give the compute figures or pricing.

TenK10K benchmark (28 immune cell types): share of causal variants identified, as reported by DeepMind (data as of )
ModelCausal variants identified
AlphaGenome 230%
Original AlphaGenome fine-tuned on single-cell data16%

FAQ

What does AlphaGenome 2 do?

It is a DNA sequence model that predicts how genetic variants affect specific human cell types and states, one condition at a time.

Can it be used in clinical diagnosis?

No. DeepMind says it is for research only and has not been validated or approved for clinical or diagnostic use.

Who can access it?

Non-commercial researchers via the AlphaGenome API; commercial research has preliminary private access.

Does it replace the original AlphaGenome?

Not in the short term, according to The Verge, because its predictions are more computationally costly.

Sources

  1. AlphaGenome 2: Deciphering how cells read DNA to better understand disease, Google DeepMind, (primary source)
  2. Google’s AlphaGenome 2 zooms in on the cellular roots of disease, The Verge,
  3. DeepMind's ultra-powerful AI for genome mapping wows scientists, Nature,

Plain text version (Markdown)