DeepMind has unveiled AlphaGenome 2, a next-generation DNA sequence model designed to predict how genetic variants impact specific human cell types and states. Announced on November 11, 2026, the model represents a significant advancement in bridging single-cell biology and human genetics, offering researchers a powerful tool to uncover the cellular mechanisms driving disease.
Enhanced Precision in Genetic Variant Prediction
AlphaGenome 2 builds on the foundation of its predecessor by moving beyond broad tissue averages to focus on individual cell types and states. This redesign allows the model to predict variant effects in disease-relevant cells that broad-tissue models might overlook. According to DeepMind, AlphaGenome 2 identifies nearly twice as many variants that alter gene expression in cells compared to the original AlphaGenome model.
The model was trained on over 100 million individual cell profiles across 68 tissues, leveraging both bulk and single-cell data from human and mouse samples. This extensive training enables AlphaGenome 2 to transform static cell atlases into a predictive engine for human genetics, offering insights into specialized cells such as heart muscle cells and immune cells.
Conditional Model Architecture
A key innovation in AlphaGenome 2 is its conditional model architecture. Unlike the original model, which predicted properties across thousands of experiments simultaneously, AlphaGenome 2 focuses on one biological condition at a time. This approach allows the model to learn how the same DNA sequence is interpreted differently across various cell types and states.
Additionally, the model supports cross-modality imputation, enabling it to predict unmeasured properties such as chromatin accessibility from single-cell gene activity data. This feature helps researchers prioritize experimental efforts and explore new cell types and states without retraining the model.
Improved Disease Variant Identification
AlphaGenome 2 has demonstrated superior performance in identifying disease-relevant genetic variants. For example, in the TenK10K dataset, which maps 28 primary immune cell types, the model identified 30% of causal variants and 43% of directional gene effects—nearly double the rates achieved by the original AlphaGenome model fine-tuned on single-cell data. This precision aids in uncovering overlooked candidate variants for further research.
In a recent study, AlphaGenome 2 computationally reproduced findings that a specific genetic variant in a regulatory switch drives inflammatory bowel disease (IBD) by overactivating the ETS2 gene in gut macrophages. The model pinpointed the causal variant and demonstrated its selective impact on gene expression in these immune cells.
Availability and Research Applications
AlphaGenome 2 is now available for non-commercial research through the AlphaGenome API and is in private preview for commercial research. By providing a finer-grained view of genetic variant effects, the model is expected to accelerate discoveries in rare diseases, complex traits, and fundamental biology, offering a transformative tool for the scientific community.
