{"schema":"sanctity-signal/0.2","record_type":"signal","id":"deepmind-alphagenome2-cells-e087","domain":"science","drop_id":"signal-e087","occurred_at":"2026-11-11","how_to_use":"This is a structured record of one real-world action and its competing interpretations, from Sanctity Signal. To build content from it (a reel script, a blog, a thread): open on the hook, ground the viewer with plain and observed, walk the dated trail, present the interpretations as live tension between named readings, and close on the before/after shift and what_to_watch. Hard rules: items under established are fact; items under unknown are never presented as fact; the evidence status (verified or reported) is stated, not dropped; sources are cited.","actor":"Google DeepMind","action_type":"publication of AlphaGenome 2, a conditional DNA model predicting variant effects per cell type, opened free for non-commercial research and placed in private preview for commercial use","observed":"Google DeepMind published \"AlphaGenome 2: Predicting how DNA shapes cells and disease\" at deepmind.google on November 11, 2026, which we read in full. AlphaGenome 2 is a conditional DNA sequence model: it takes a 1 million letter DNA sequence plus a target condition, a tissue, cell type or cell state, and predicts effects for that condition. Outputs include gene expression, histone modifications, chromatin accessibility and DNA methylation. It can predict a missing measurement, such as chromatin accessibility, from a single measured readout. Training used bulk and single-cell data from human and mouse samples: the post gives over 100 million individual cell profiles in one place and over 107 million in another, across 68 tissues. Results: on the TenK10K dataset of 28 primary immune cell types it identified causal variants in 30% of cases against 16% for the original AlphaGenome fine-tuned on single-cell data and other specialised tools, and found directional gene effects in 43% of cases against 23%. DeepMind says it identifies nearly twice as many variants that change gene expression in specific cells as the previous state of the art. Two disease examples are given, both reproducing published findings: a distant regulatory variant raising the ETS2 gene in gut macrophages in inflammatory bowel disease, and a severe COVID-19 linked variant disrupting gene activity more strongly in activated T cells. Access: free through the AlphaGenome application programming interface for non-commercial research, private preview for commercial research, precomputed predictions for trait-associated variants across hundreds of conditions, Colab tutorials, and a call for use cases by November 30. DeepMind says it is for research use only and has not been validated for clinical or diagnostic use, and names the Wellcome Sanger Institute as a collaborator. Sarah Teichmann of the University of Cambridge and Peter Donnelly, chief executive of Genomics PLC, are quoted; Donnelly says his company plans to incorporate it into its Mystra platform. Not published: the price of commercial access, the compute used, the model size, or any laboratory validation of a prediction the model made first.","corrections":[],"evidence":{"status":"verified","note":null,"sources":[{"role":"primary","label":"Google DeepMind: AlphaGenome 2, Predicting how DNA shapes cells and disease, deepmind.google, Nov 11, 2026","url":"https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/"},{"role":"corrob","label":"The Verge: Google DeepMind's AlphaGenome 2 genetic AI model, Nov 11, 2026","url":"https://www.theverge.com/ai-artificial-intelligence/1031524/google-deepmind-alphagenome-2-genetic-ai-model"},{"role":"corrob","label":"OODAloop: Google DeepMind unveils AlphaGenome 2 to bridge single-cell biology and human genetics, Nov 11, 2026","url":"https://oodaloop.com/briefs/technology/google-deepmind-unveils-alphagenome-2-to-bridge-single-cell-biology-and-human-genetics/"}]},"strength":{"cost":"Medium","commitment":"Medium","reversibility":"Moderate","persistence":"Years","rationale":"Cost is medium: training on over 107 million cell profiles is expensive, and giving the result away free to academics forgoes revenue, but this is a research release rather than a capital commitment. Commitment is medium because published numbers invite replication and a named collaborator and a named customer have attached themselves to it. Reversibility is moderate: an application programming interface can be closed or priced, as the private preview for commercial users already signals. Persistence is years, because precomputed predictions and published benchmarks stay useful, and a free tool academics build on is hard to withdraw. We do not score it higher: both disease examples reproduce findings already in the literature."},"shift":{"prior":"AI reads the genome in bulk, averaged across a whole tissue.","action":"Google DeepMind published \"AlphaGenome 2: Predicting how DNA shapes cells and disease\" at deepmind.google on November 11, 2026, which we read in full. AlphaGenome 2 is a conditional DNA sequence model: it takes a 1 million letter DNA sequence plus a target condition, a tissue, cell type or cell state, and predicts effects for that condition. Outputs include gene expression, histone modifications, chromatin accessibility and DNA methylation. It can predict a missing measurement, such as chromatin accessibility, from a single measured readout. Training used bulk and single-cell data from human and mouse samples: the post gives over 100 million individual cell profiles in one place and over 107 million in another, across 68 tissues. Results: on the TenK10K dataset of 28 primary immune cell types it identified causal variants in 30% of cases against 16% for the original AlphaGenome fine-tuned on single-cell data and other specialised tools, and found directional gene effects in 43% of cases against 23%. DeepMind says it identifies nearly twice as many variants that change gene expression in specific cells as the previous state of the art. Two disease examples are given, both reproducing published findings: a distant regulatory variant raising the ETS2 gene in gut macrophages in inflammatory bowel disease, and a severe COVID-19 linked variant disrupting gene activity more strongly in activated T cells. Access: free through the AlphaGenome application programming interface for non-commercial research, private preview for commercial research, precomputed predictions for trait-associated variants across hundreds of conditions, Colab tutorials, and a call for use cases by November 30. DeepMind says it is for research use only and has not been validated for clinical or diagnostic use, and names the Wellcome Sanger Institute as a collaborator. Sarah Teichmann of the University of Cambridge and Peter Donnelly, chief executive of Genomics PLC, are quoted; Donnelly says his company plans to incorporate it into its Mystra platform. Not published: the price of commercial access, the compute used, the model size, or any laboratory validation of a prediction the model made first.","update":"It can now answer which cell type a variant acts in."},"inference":{"established":["On TenK10K immune data it found causal variants in 30% of cases versus 16%.","It found directional gene effects in 43% of cases versus 23%.","Non-commercial research access is open now through the AlphaGenome interface."],"unknown":["Any clinical result. Google says it is not validated for diagnosis.","A new drug or treatment. Both disease examples reproduce published findings.","Commercial terms. Business access sits in an unpriced private preview."]},"narrative":{"hook":"Google released a DNA model that reads genes cell by cell, free for research.","plain":"On November 11 Google DeepMind published AlphaGenome 2, trained on over 107 million single cells across 68 tissues. Caveat: it is for research only and is not validated for clinical or diagnostic use.","story":["Most disease-linked DNA changes sit outside genes, and the hard question is which cell they act in. On immune cell data the model found the causal variant in 30% of cases against 16% for the previous version. Google gave it to academics free and put commercial access behind an unpriced private preview."],"trail":[{"d":"JUN 2025","t":"Google DeepMind releases the first AlphaGenome, a model predicting how changes in a DNA sequence affect gene regulation, free to non-commercial researchers through an application programming interface."},{"d":"SEP 2026","t":"DeepMind publishes AlphaGenome Atlas, a precomputed map of molecular predictions for every possible single-letter change in the human genome."},{"d":"NOV 6","t":"DeepMind publishes a machine-checked result cutting the cost of factoring 1024-bit keys. Edition 080."},{"d":"NOV 10","t":"DeepMind publishes Gemini Robotics 2.5 with a 96% kitting result from active Mercedes-Benz lines in Hungary and Germany. Edition 086."},{"d":"NOV 11","t":"AlphaGenome 2 is published: conditional on cell type, trained on over 107 million cell profiles across 68 tissues, free for non-commercial research and in private preview for commercial use."},{"d":"Reacts","r":true,"t":"Genomics PLC, which says it will put the model inside its own platform; the Wellcome Sanger Institute, named as a collaborator; drug discovery teams that pay for variant interpretation today; medicines regulators, who have no approved route for a model like this; and the academic groups whose single-cell tools it just outscored."}],"glossary":[{"term":"Variant","def":"A difference between one person's DNA and the reference sequence. Most sit outside genes and act by changing how nearby genes are switched on."},{"term":"Gene expression","def":"How strongly a gene is switched on in a cell. The same DNA gives different expression in a nerve cell and a skin cell."},{"term":"Single-cell data","def":"Measurements taken from individual cells rather than averaged across a whole tissue sample."},{"term":"Causal variant","def":"The one DNA change actually responsible for an effect, among many nearby changes that are only correlated with it."}]},"judgment":{"prompt":"Which explanation do you believe?","interpretations":[{"id":"a","label":"Biology is the next frontier","text":"The labs are moving from language to the cell, where the hard data is.","abstain":false},{"id":"b","label":"A research tool, not medicine","text":"Doubling variant hits is useful. It changes no treatment by itself.","abstain":false},{"id":"c","label":"A move to own genomics","text":"Free to academics, private preview to industry, is how a platform is seeded.","abstain":false},{"id":"d","label":"The normal annual model refresh","text":"AlphaGenome shipped last year, an atlas in September, version 2 now.","abstain":false},{"id":"e","label":"Too early to call.","text":"The evidence cannot yet separate these.","abstain":true}],"distribution":{"status":"held back until you have made your call","calls":null}},"story_thread":null,"what_to_watch":[{"linked_interpretation":"a","direction":"strengthens","observable_event":"An independent group reproduces the 30% causal-variant result on data Google did not train on, or a peer-reviewed paper reports a variant found first by AlphaGenome 2 and later confirmed in the laboratory, by December 31, 2027.","window_end":"2027-12-31","state":"unresolved"},{"linked_interpretation":"a","direction":"weakens","observable_event":"A published benchmark finds AlphaGenome 2 no better than existing single-cell tools outside its training distribution, by December 31, 2027.","window_end":"2027-12-31","state":"unresolved"},{"linked_interpretation":"c","direction":"strengthens","observable_event":"Google prices commercial access to AlphaGenome 2, or names a paying pharmaceutical customer, by September 30, 2027.","window_end":"2027-09-30","state":"unresolved"}]}