# Nature reports AlphaGenome 2 replicating aspects of an IBD discovery [Canonical reading](https://escapemeridian.com/readings/nature-reports-alphagenome-2-replicating-aspects-of-an-ibd-605a8330) Nature reports that Google DeepMind’s AlphaGenome 2 model replicated aspects of a 2024 finding about ETS2, a gene linked to inflammatory bowel disease (IBD), without laboratory experiments. The original researchers used gene-editing and other molecular biology techniques to identify the gene and potential treatments. James Lee, who led that study, says the model reached the answer in 30 minutes, compared with a year and a half of experiments, and calls the result “reassuring from our perspective.” The reported change is cell-type specificity: AlphaGenome 2 predicts variant effects across different cell types, a key distinction because non-coding DNA sequences can be active in particular cells. The earlier AlphaGenome 1 predictions did not specify the cell type in which a variant was likely to act; Lee says that limitation made them not hugely useful to his team. Google DeepMind says it trained the new model on measurements from individual cells across tissues, rather than bulk samples of mixed cells. Researchers can also supply new single-cell gene activity data, such as data from cells in a disease state, to generate predictions. The article presents this as a possible way to identify disease-related variants more quickly, not as a demonstrated clinical benefit. The model is described in a preprint on the company’s website. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Nature · Article Original source date: Nov 11, 2026 (2026-11-11T00:00:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-12T04:47:44.355Z. First observed: 2026-11-11T16:21:29.131Z. Retrieved: 2026-11-12T04:46:49.817Z. - [DeepMind's ultra-powerful AI for genome mapping wows scientists](https://www.nature.com/articles/d41586-026-03673-0) ## Limits The article reports replication of aspects of the finding, not a full replication of the 2024 study. The article describes the model in a preprint on the company’s website; it does not report an independent evaluation of the model’s predictions on new disease-state data. The article says predictions could help identify disease-related variants more quickly, but does not report that this benefit has been demonstrated. ## Open question How well do predictions made with new single-cell gene activity data identify variants underlying disease? ## Source-stated Nature reports that AlphaGenome 2 replicated aspects of a 2024 finding linking ETS2 to inflammatory bowel disease without laboratory experiments. > Now, researchers at Google DeepMind in London have replicated aspects of the finding using an AI model, and they have done this without lifting a pipette. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source-stated James Lee says AlphaGenome 2 reached the answer in 30 minutes, compared with a year and a half of experiments, and describes the result as reassuring. > “They’ve managed to figure out the answer in 30 minutes, and it took us a year and a half of experiments,” says James Lee, an immunologist and geneticist at the Francis Crick Institute in London who led the 2024 study. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source-stated AlphaGenome 2 predicts variant effects across different cell types, addressing a limitation Lee identifies in AlphaGenome 1 predictions. > These and other predictions were made with a tool called AlphaGenome 1 and came with a major caveat: they did not specify the precise kind of cell – be it antibody-making B cells or pancreatic beta-cells – in which a variant was likely to act. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source-stated Google DeepMind says AlphaGenome 2 was trained on measurements from individual cells across 68 tissues, while AlphaGenome 1 used bulk samples of heterogeneous cells. > To develop an AI model that could decipher variant effects in diverse cells, Avsec’s team trained AlphaGenome 2 on gene activity and other measurements made in more than 100 million individual cells, across 68 tissues. AlphaGenome 1, by contrast, was fed data collected in ‘bulk’ samples of hete - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source-stated The article says researchers can provide new single-cell gene activity data to generate predictions, which could help identify disease-related variants more quickly. > This could help researchers more quickly identify variants underlying disease, says Gayoso. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Escape Meridian interpretation The article describes the model in a preprint on Google DeepMind’s website; the reported replication is of aspects of the earlier finding. > Their model is called AlphaGenome 2. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source and contributor context Nature · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media declared by the original publisher; availability checked separately from the editorial reading. - [image · Nature](https://media.nature.com/w140h79/springer-static/image/art%3A10.1038%2Fs41586-026-11204-0/MediaObjects/41586_2026_11204_Fig1_HTML.png) ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Nature Nature publishes research, news, and commentary across the sciences. - [Explore this context](/entities/nature) - [Reference](https://www.nature.com/) - [Reference](https://www.nature.com/nature.rss) ## Google DeepMind Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold. - [Explore this context](/entities/google-deepmind) - [Reference](https://deepmind.google/blog/) - [Reference](https://deepmind.google/science/alphafold/) ## Editorial provenance Version 1. Publication ID: 7ed6d5f0-3781-47d8-8db1-5ecbd6c0a938. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T04:47:44.355Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 5448384ca939bf160fce07c02be2b747d5a7b15ba7e777ea846dc682b819bb82. Source text hash: 7c1bf6759082654967dad3c318b573543cdbad37c74b1ce24b0f6c67c7530ee2. ## Revision and source-check history ### Version 1 — 2026-11-12T04:47:44.355Z Nature reports that Google DeepMind’s AlphaGenome 2 model replicated aspects of a 2024 finding about ETS2, a gene linked to inflammatory bowel disease (IBD), without laboratory experiments. The original researchers used gene-editing and other molecular biology techniques to identify the gene and potential treatments. James Lee, who led that study, says the model reached the answer in 30 minutes, compared with a year and a half of experiments, and calls the result “reassuring from our perspective.” The reported change is cell-type specificity: AlphaGenome 2 predicts variant effects across different cell types, a key distinction because non-coding DNA sequences can be active in particular cells. The earlier AlphaGenome 1 predictions did not specify the cell type in which a variant was likely to act; Lee says that limitation made them not hugely useful to his team. Google DeepMind says it trained the new model on measurements from individual cells across tissues, rather than bulk samples of mixed cells. Researchers can also supply new single-cell gene activity data, such as data from cells in a disease state, to generate predictions. The article presents this as a possible way to identify disease-related variants more quickly, not as a demonstrated clinical benefit. The model is described in a preprint on the company’s website. ## Original source Nature · Article Original source date: Nov 11, 2026 (2026-11-11T00:00:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-12T04:47:44.355Z. First observed: 2026-11-11T16:21:29.131Z. Retrieved: 2026-11-12T04:46:49.817Z. - [DeepMind's ultra-powerful AI for genome mapping wows scientists](https://www.nature.com/articles/d41586-026-03673-0) ## Limits The article reports replication of aspects of the finding, not a full replication of the 2024 study. The article describes the model in a preprint on the company’s website; it does not report an independent evaluation of the model’s predictions on new disease-state data. The article says predictions could help identify disease-related variants more quickly, but does not report that this benefit has been demonstrated. ## Open question How well do predictions made with new single-cell gene activity data identify variants underlying disease? ## Source-stated Nature reports that AlphaGenome 2 replicated aspects of a 2024 finding linking ETS2 to inflammatory bowel disease without laboratory experiments. > Now, researchers at Google DeepMind in London have replicated aspects of the finding using an AI model, and they have done this without lifting a pipette. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source-stated James Lee says AlphaGenome 2 reached the answer in 30 minutes, compared with a year and a half of experiments, and describes the result as reassuring. > “They’ve managed to figure out the answer in 30 minutes, and it took us a year and a half of experiments,” says James Lee, an immunologist and geneticist at the Francis Crick Institute in London who led the 2024 study. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source-stated AlphaGenome 2 predicts variant effects across different cell types, addressing a limitation Lee identifies in AlphaGenome 1 predictions. > These and other predictions were made with a tool called AlphaGenome 1 and came with a major caveat: they did not specify the precise kind of cell – be it antibody-making B cells or pancreatic beta-cells – in which a variant was likely to act. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source-stated Google DeepMind says AlphaGenome 2 was trained on measurements from individual cells across 68 tissues, while AlphaGenome 1 used bulk samples of heterogeneous cells. > To develop an AI model that could decipher variant effects in diverse cells, Avsec’s team trained AlphaGenome 2 on gene activity and other measurements made in more than 100 million individual cells, across 68 tissues. AlphaGenome 1, by contrast, was fed data collected in ‘bulk’ samples of hete - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source-stated The article says researchers can provide new single-cell gene activity data to generate predictions, which could help identify disease-related variants more quickly. > This could help researchers more quickly identify variants underlying disease, says Gayoso. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Escape Meridian interpretation The article describes the model in a preprint on Google DeepMind’s website; the reported replication is of aspects of the earlier finding. > Their model is called AlphaGenome 2. - [Supporting source](https://www.nature.com/articles/d41586-026-03673-0) ## Source and contributor context Nature · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media declared by the original publisher; availability checked separately from the editorial reading. - [image · Nature](https://media.nature.com/w140h79/springer-static/image/art%3A10.1038%2Fs41586-026-11204-0/MediaObjects/41586_2026_11204_Fig1_HTML.png) ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Nature Nature publishes research, news, and commentary across the sciences. - [Explore this context](/entities/nature) - [Reference](https://www.nature.com/) - [Reference](https://www.nature.com/nature.rss) ## Google DeepMind Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold. - [Explore this context](/entities/google-deepmind) - [Reference](https://deepmind.google/blog/) - [Reference](https://deepmind.google/science/alphafold/) ## Editorial provenance Version 1. Publication ID: 7ed6d5f0-3781-47d8-8db1-5ecbd6c0a938. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T04:47:44.355Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 5448384ca939bf160fce07c02be2b747d5a7b15ba7e777ea846dc682b819bb82. Source text hash: 7c1bf6759082654967dad3c318b573543cdbad37c74b1ce24b0f6c67c7530ee2. ### uncertain — 2026-11-12T04:47:44.355Z — anthropic/claude-opus-5.5 Publication: 7ed6d5f0-3781-47d8-8db1-5ecbd6c0a938 attribution — pass: Nature is credited as the reporter, and quotes are correctly assigned to James Lee and Adam Gayoso. The phrase "Google DeepMind says" for the training data is slightly looser than the source, where Nature narrates what Avsec's team did. This does not misattribute a claim. > Avsec’s team trained AlphaGenome 2 on gene activity and other measurements made in more than 100 million individual cells, across 68 tissues summary — uncertain: The summary is faithful, but central terms go unexplained. A non-specialist is not told what a DNA variant is or that non-coding DNA is the roughly 98% of the genome that does not code for proteins. The core function also goes unstated: the source says AlphaGenome predicts the molecular effects of non-coding sequences, such as effects on other genes' expression. Minimal repair: add one plain-language sentence saying what the model predicts and what non-coding DNA is. Suggest term-library tooltips for non-coding DNA, variant and single-cell data. > an AI tool that predicts the molecular effects of ‘non-coding’ DNA sequences, such as on the expression of other genes. These sequences comprise 98% of the human genome date — pass: originalPublished is 2026-11-11, which matches the article's statement that the model was "announced today" and the 11 NOV 26 dates shown alongside it. There is no sign of an earlier first publication. > The AlphaGenome 2 model, which was announced today support — uncertain: Most claims are source-stated and well hedged; the limitations correctly avoid presenting the benefit as demonstrated. However, claim 6 asserts that the model is described in a preprint on Google DeepMind's website. Its excerpt ("Their model is called AlphaGenome 2.") does not support that statement. The supporting text exists elsewhere in the source and should replace the excerpt. > co-author of a preprint describing the model on the company’s website media — pass: The selected media is the verified direct link to the original Nature article, which is the most relevant way to access the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/605a8330e60883f354fa0cdae81e68ef549e3db8f5be574440e808b09ab970d0/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/nature-reports-alphagenome-2-replicating-aspects-of-an-ibd-605a8330.json)