Understand this story
Google DeepMind launches AlphaGenome 2 model. According to an Ask a Scientist interview published on the Google Blog on November 17, 2026, Google DeepMind research scientist Žiga Avsec introduced how AlphaGenome and AlphaGenome Atlas work: AlphaGenome can take in a local DNA sequence of 1 million base pairs and make thousands of predictions; the team used it to predict all 9 billion single-letter variants, forming the AlphaGenome Atlas dataset, and produced a single AVI score to help researchers rank variants. The article notes that AlphaGenome was released in June 2025 and AlphaGenome Atlas was launched this September; the latest model, AlphaGenome 2, addresses the limitation of cell-type specificity by moving from tissue-level averages to predicting how variants affect individual cell types. The interview also mentions that Gareth Hawkes of the University of Exeter used Atlas to screen 1 billion variants and identify 19 variant groups associated with BMI, and that Laura Covill of the Broad Institute used the AVI score in rare disease research to find key variants associated with neurodevelopmental disorders.
AI-generated from public reports
Latest development · Report
Google DeepMind researcher explains how AlphaGenome and the AlphaGenome Atlas workDevelopments
1 confirmed developments
- Source:Google Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 17
Google BlogFirst-party
Google DeepMind researcher explains how AlphaGenome and the AlphaGenome Atlas work
Google DeepMind research scientist Žiga Avsec, in an Ask a Scientist interview, explained how AlphaGenome and the AlphaGenome Atlas work: AlphaGenome can take in a local DNA sequence of 1 million base pairs and make thousands of predictions, and the team used it to predict all 9 billion single-letter variants, forming the AlphaGenome Atlas dataset and producing a single AVI score to help researchers rank variants. AlphaGenome was released in June 2025, and the AlphaGenome Atlas was launched this September; the latest model, AlphaGenome 2, addresses the limitation of cell-type specificity by moving from tissue-level averages to predicting how variants affect individual cell types. The article also mentions that Gareth Hawkes of the University of Exeter used Atlas to screen 1 billion variants down to 19 variant groups associated with BMI, and that Laura Covill of the Broad Institute used AVI scores in rare disease research to find key variants associated with neurodevelopmental disorders.
AI Trends