Independent newsroom

Thursday, November 12, 2026

The File

Every claim traces to a primary source. Uncertainty is labeled. Corrections are public.

Cells · Thursday, November 12, 2026

DeepMind says its new model reads DNA one cell type at a time

The File desk · Nov. 12, 2026, 6:41 a.m. ET

Partially confirmed1 primary sourcesAdded 6:41 a.m. ET, Nov. 12

Status

Partially confirmed. Google DeepMind's November 11, 2026 post announces AlphaGenome 2 and says it is for research, not for clinical or diagnostic use. The performance figures are the company's. The post does not say a journal has reviewed the model.
  • Confirmed
    The post is dated November 11, 2026. It says AlphaGenome 2 predicts how DNA variants affect gene activity in a specific human cell type, is available for non-commercial research, and is not validated for clinical or diagnostic use.
  • Partially confirmed
    The training count, the TenK10K comparison, and the ETS2 reproduction are DeepMind's. This desk did not reproduce them.
Google DeepMind said on Wednesday, November 11, 2026, that it is releasing AlphaGenome 2, a model that predicts how a change in a person's DNA changes gene activity in one kind of human cell.1 The company says every cell carries the same DNA, yet a heart muscle cell and an immune cell switch different genes on or off.1
The change the model looks at is called a genetic variant, a difference in the DNA letters between people, sometimes a single letter. AlphaGenome 2 reads a million-letter stretch of DNA and is built to predict the difference in gene expression, meaning how strongly a gene is switched on, for one condition at a time: a tissue, a cell type, or a cell state.1
DeepMind says it trained the model on over 107 million individual cell profiles.1 The company tested the model on TenK10K, a map of immune cells, and says it identified causal variants 30 percent of the time.1 DeepMind says its earlier AlphaGenome model, after it was adjusted on single-cell data, did so 16 percent of the time.1
The company also says the model reproduced, in a computer run, a published finding that a variant overactivates the ETS2 gene in gut macrophages in inflammatory bowel disease, which the post abbreviates as IBD.1 The model adds a prediction for DNA methylation, a chemical mark on DNA.1 AlphaGenome 2 is available for non-commercial research through the AlphaGenome API, and in a private preview for commercial research.1
Why it matters (Estimate). Estimate. A researcher studying how the same DNA change acts in one cell type and not another can query the API for non-commercial work now. It is not a test a clinic can run. The 30 percent and 16 percent figures, and the claim that the model recovered the ETS2 finding, are DeepMind's.
The post carries its own warning: "Please note that AlphaGenome and AlphaGenome 2 are intended for research use only and have not been validated or approved for clinical or diagnostic purposes."1
How we know

Notes

  1. 1.Google DeepMind, "AlphaGenome 2: Deciphering how cells read DNA to better understand disease," visible line November 11, 2026. https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/ The page's structured data lists datePublished 2026-11-11T15:00:00+00:00, which is 10:00 a.m. ET. The visible byline does not print an hour. Authors on the page: Adam Gayoso, Natasha Latysheva, Pushmeet Kohli and Žiga Avsec. "Today, we are announcing AlphaGenome 2, our next-generation model that predicts how DNA sequences and their variations impact specific human cell types." "Every cell in our body carries the same DNA blueprint, yet each cell type and state—like a heart muscle cell or an immune cell fighting an infection—switches different genes on or off to do its job." "AlphaGenome 2 is a conditional model that predicts DNA effects for one condition - defined as a tissue, cell type or state - at a time." It uses "the million-letter DNA sequence." "We trained AlphaGenome 2 on both bulk and single-cell data from human and mouse samples, including over 107 million individual cell profiles." A separate sentence on the page, not used as the body's count: "Trained on more than 100 million individual cell profiles across 68 tissues." "On the TenK10K dataset, a large single-cell map of 28 primary immune cell types, 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 specialized tools." The 43 percent and 23 percent comparison is DeepMind's and is not in the story body. "A recent landmark study showed that a variant in a distant regulatory switch drives IBD by overactivating the ETS2 gene in gut immune cells (macrophages)." The post says AlphaGenome 2 "reproduced this finding computationally." The post links https://www.nature.com/articles/s41586-024-07501-1 . This desk did not open that paper. The model "also" predicts "DNA methylation, a widely requested readout." "AlphaGenome 2 is available today for non-commercial research via the AlphaGenome API, and in private preview for commercial research." "Please note that AlphaGenome and AlphaGenome 2 are intended for research use only and have not been validated or approved for clinical or diagnostic purposes." This desk found no earlier AlphaGenome story in the edition files. The post says DeepMind launched AlphaGenome last year. Retrieved 6:25 a.m. ET, Nov. 12.

What is still unknown or disputed

  • Whether a journal has reviewed AlphaGenome 2. The post does not say it has.
  • How the model performs outside DeepMind's account. This desk did not run it.
  • A clock on the visible byline. The page prints November 11, 2026, and does not print an hour there.

Primary sources

Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.

  1. Source 1

    AlphaGenome 2: Deciphering how cells read DNA to better understand disease

    Google DeepMind · November 11, 2026

    DeepMind said on November 11, 2026, that AlphaGenome 2 predicts how DNA variants change gene activity in a specific human cell type. It is for non-commercial research now, and the post says it is not validated for clinical or diagnostic use.

    https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/