Google DeepMind Releases AlphaGenome 2 for Disease Research
Google DeepMind released its next-generation DNA-sequence prediction model, AlphaGenome 2. [[fact:fact-la19xk0dh]] The article says the model computationally reproduced a key finding by James Lee’s team on inflammatory bowel disease in half an hour. [[fact:fact-caxyo9f35]] Its training data cover 68 tissue types and more than 107 million human and mouse cells. [[fact:fact-weje1zus6]] In the TenK10K test covering 28 immune-cell types, AlphaGenome 2 achieved 30% recall for key causal variants, compared with 16% for the previous generation after single-cell-data fine-tuning. [[fact:fact-c8wzlc3av]] Its ability to predict whether a variant would increase or decrease gene activity rose from 23% to 43%. [[fact:fact-9553ax0d5]] Google is using the model to study complex diseases including type 2 diabetes, screening DNA-variant clues and identifying related cells and genes; the model is also available for non-commercial research. [[fact:fact-2q0ovqija]][[fact:fact-q1z6ngqji]]