Google DeepMind's AlphaGenome 2 release

Google DeepMind released AlphaGenome 2 on November 11, 2026, a model that takes both a DNA sequence and a target cell type or state as input, predicting regulatory effects one condition at a time. The model accepts a one-million-letter DNA sequence plus a cell type and predicts that cell's regulatory response. Trained on over 107 million individual cell profiles across 68 tissues, it identifies nearly twice as many causal variants as the previous AlphaGenome model, 30% versus 16% on 28 immune cell types from the TenK10K dataset, and predicts effect direction at 43% versus 23% for the prior best. Those benchmark numbers come from DeepMind's own evaluation; the model is for research use and is not validated for clinical or diagnostic use. DeepMind precomputed predictions for trait-linked variants across hundreds of conditions, covering traits from height to type 2 diabetes. The model reproduced a published finding that a specific variant drives inflammatory bowel disease by overactivating the ETS2 gene specifically in gut macrophages, leaving other cell types and genes largely unaffected. Its ability to accept single-cell RNA-seq data as conditioning input lets it generalize to new, unseen cell types and disease states without retraining. The AlphaGenome API is free for non-commercial use via GitHub; API outputs cannot be used to train other machine learning models. Commercial access is available via Google Cloud. DeepMind plans to integrate AlphaGenome 2 into its Antigravity agentic science platform alongside Gemini 4. Separately, researchers have applied the original AlphaGenome model to Alzheimer's disease research, using it to nominate candidate causal genes in a GWAS of argyrophilic grain disease and finding that one AD risk variant may alter transcription and splicing of the Fc receptor gene FCER1G, linking immune signaling to AD risk.