DeepMind Unveils AlphaGenome 2 to Predict Disease-Linked Genetic Variants in Specific Cell Types

AlphaGenome 2, a next-generation model from Google DeepMind, predicts how genetic variations affect gene expression across hundreds of human cell types with significantly improved accuracy. Trained on over 100 million single-cell profiles from 68 tissues, it identifies nearly twice as many disease-relevant variants compared to its predecessor, enabling researchers to study cell-specific effects in conditions previously overlooked by broad-tissue models. The model is now available via API for non-commercial research and in private preview for commercial use.
On November 11, 2026, Google DeepMind unveiled AlphaGenome 2, a next-generation artificial intelligence model designed to predict how genetic variations influence gene expression across hundreds of human cell types with unprecedented precision. Trained on over 107 million single-cell profiles from 68 tissues—spanning both human and mouse samples—the model identifies nearly twice as many disease-relevant variants compared to its predecessor, AlphaGenome, particularly in cell types previously obscured by broad-tissue averaging. Unlike the earlier model, which generated parallel predictions across multiple tissue types simultaneously, AlphaGenome 2 employs a conditional architecture that processes one biological condition—such as a specific cell type or state—at a time, enabling fine-grained analysis of how DNA sequences are interpreted differently in distinct cellular environments. This shift allows it to predict variant effects even in novel or rare cell states, including those derived from a researcher’s own single-cell experiment via an integrated cell encoder. The model supports key molecular readouts like gene expression, chromatin accessibility, histone modifications, and DNA methylation, with the ability to impute missing data across modalities—such as predicting chromatin state from gene activity alone.
The development builds on prior work in human genetics and single-cell biology, including the Human Cell Atlas initiative and DeepMind’s earlier AlphaGenome and AlphaGenome Atlas platforms. These tools were instrumental in mapping variant effects across broad tissues but struggled to capture cell-type-specific regulatory dynamics. By leveraging advances in deep learning architecture and massive-scale biological data, AlphaGenome 2 overcomes these limitations. Key figures like Professor Sarah Teichmann of the University of Cambridge, a co-founder of the Human Cell Atlas, have praised its ability to adapt to new single-cell data without retraining. The model is now accessible via API for non-commercial research and in private preview for commercial use, signaling a shift toward democratizing high-resolution genetic prediction tools. What remains unresolved is how widely the model will be adopted across diverse research domains, whether it will uncover clinically actionable variants in complex diseases, and how regulatory frameworks may evolve to govern its use in medicine.
AlphaGenome 2 could significantly accelerate discovery in precision medicine by enabling researchers to pinpoint disease-driving genetic variants in specific cell types—such as immune cells or neurons—where they might otherwise go undetected. This granular insight may improve the identification of therapeutic targets for conditions like autoimmune disorders, neurodegenerative diseases, and rare genetic syndromes, potentially reducing reliance on trial-and-error drug development. The model’s ability to impute missing experimental data may also lower costs and time in lab workflows, particularly for researchers with limited access to multi-omics platforms. In the broader biotech ecosystem, this could spur innovation in computational biology tools and increase demand for single-cell datasets, possibly reshaping how pharmaceutical companies prioritize drug candidates. A comparable precedent is the impact of AlphaFold on structural biology—once a niche tool, it became foundational across drug discovery and protein engineering. Similarly, AlphaGenome 2 may become a standard reference for interpreting genetic variation, though its real-world utility will depend on validation in clinical settings. What to watch for next includes whether the model’s predictions lead to new diagnostic markers or therapies, how quickly commercial applications emerge, and whether ethical concerns around data privacy and algorithmic bias in genomics will surface as usage expands.