> ## Content Index
> Fetch the complete content index at: https://soundbytes.news/llms.txt
> Use this file to discover other available public pages before exploring further.

# Google DeepMind's new model predicts how DNA changes act in chosen cell types
- URL: https://soundbytes.news/google-deepminds-new-model-predicts-how-dna-changes-act-in-chosen-cell-types/
- Published: 2026-11-11T15:21:09.000Z
- Updated: 2026-11-18T08:49:46.000Z
- Description: Google DeepMind says its updated model takes a DNA sequence and a chosen cell type and predicts how DNA changes affect genes there. It is open to non-commercial researchers and is for research use only.
- Author: Echo
- Tags: Science and discovery

A computer model has reproduced an earlier finding on inflammatory bowel disease, [Google DeepMind says](https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/?ref=soundbytes.news). It singled out a DNA variant and showed that it boosts one gene mainly in gut immune cells. A variant is a spot where one person's DNA differs from another's.

The model is called AlphaGenome 2\. An earlier study had linked the variant to the disease through over-activation of a gene called ETS2 in gut immune cells called macrophages. In a post dated 11 November 2026, DeepMind says the model reproduced that finding on a computer, with other cell types and genes largely unaffected.

Every cell carries the same DNA, but different cell types switch different genes on or off. DeepMind's post calls predicting how small DNA differences change this an unsolved challenge.

The model takes a million-letter DNA sequence and a chosen cell type or cell state. It predicts effects for that one condition at a time.

In DeepMind's own tests on TenK10K, a single-cell map of 28 immune cell types, AlphaGenome 2 identified 30% of causal variants (DNA changes that drive an effect). The comparison figure, for the original AlphaGenome fine-tuned (given extra training) on single-cell data and other specialised tools, was 16%. For directional gene effects, the figures were 43% and 23%.

Scientists can upload gene-activity data from just a few dozen cells, and the model predicts for that new cell state without being retrained.

Professor Sarah Teichmann, co-founder of the Human Cell Atlas consortium and Professor of Stem Cell Medicine at the University of Cambridge, said in the post that AlphaGenome 2 "adapts directly to new single-cell data without retraining".

AlphaGenome 2 is available for non-commercial research through the AlphaGenome API, an online access route, and in private preview for commercial research. DeepMind says AlphaGenome and AlphaGenome 2 are intended for research use only and have not been validated or approved for clinical or diagnostic purposes.

Researchers at Charité University Medicine Berlin and the Berlin Institute for Medical Systems Biology posted a [preprint](https://www.biorxiv.org/content/10.64898/2026.11.06.764851v1?ref=soundbytes.news), a study shared before peer review, on 9 November 2026\. It tested the original AlphaGenome, not version 2.

On 2,111 pairs of genes and promoters (the stretch of DNA where a gene's reading begins), the original AlphaGenome's predictions reflected a local loss of promoter activity. They did not consistently carry that effect across the whole transcript, the gene's readout.

The authors say systematic benchmarking and lab validation remain essential before such predictions are used to prioritise or classify variants.

DeepMind says it is working with collaborators, including researchers at the Wellcome Sanger Institute, to profile more human cell types.

## Sources

- [AlphaGenome 2: Deciphering how cells read DNA to better understand disease](https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/?ref=soundbytes.news) · deepmind.google
- [Systematic promoter perturbation reveals limitations of AlphaGenome gene expression predictions | bioRxiv](https://www.biorxiv.org/content/10.64898/2026.11.06.764851v1?ref=soundbytes.news) · biorxiv.org