Google DeepMind Introduces AlphaGenome 2 to Predict How DNA Variations Affect Human Cells
Google DeepMind has introduced AlphaGenome 2, an artificial intelligence model designed to predict how DNA sequences and genetic variations influence specific human cell types. Announced on November 11, 2026, the system builds on the company's earlier genomics research by examining genetic effects at a more detailed cellular level.
AlphaGenome 2 was trained on more than 107 million individual cell profiles collected across 68 tissues, using both human and mouse biological data. The model aims to help researchers understand how genetic variations influence gene activity in different cells, potentially revealing biological mechanisms associated with diseases.
AlphaGenome 2 Moves Beyond Broad Tissue-Level Predictions
Most cells in the human body contain essentially the same DNA, but they do not use the same genes in the same way. Different cell types activate or suppress particular genes depending on their functions and biological states.
This distinction matters when investigating genetic variations. A DNA change may influence gene activity in an immune cell but have a different effect in a heart muscle cell. Models that examine broad tissue averages can overlook these differences.
AlphaGenome 2 addresses this challenge through a conditional modelling approach. Researchers can provide a DNA sequence alongside a specific tissue, cell type or biological state. The system then predicts how that sequence may influence the selected biological condition.
The model can analyse gene expression, histone modifications and chromatin accessibility. It also adds DNA methylation predictions, providing another way to study how cells regulate genetic activity.
Improved Accuracy in Identifying Disease-Related Variants
Google DeepMind reported significant improvements in the model's ability to identify genetic variations associated with changes in gene expression.
In a benchmark involving 28 primary immune cell types from the TenK10K dataset, AlphaGenome 2 identified 30% of causal variants, compared with 16% for the previous AlphaGenome model. Its performance in identifying the direction of gene effects also improved from 23% to 43%.
The researchers demonstrated another application involving inflammatory bowel disease. A genetic variation linked to the condition can increase activity of the ETS2 gene in gut macrophages, a type of immune cell. AlphaGenome 2 reproduced this biological finding computationally and identified the cell-specific effect.
These results could help researchers narrow down which genetic variations deserve further investigation rather than relying entirely on broad associations between DNA and disease.
Supporting Research Across New Cell Types
AlphaGenome 2 can also make predictions for cell types and biological states that were not included in its original training conditions. Researchers can provide new single-cell gene activity measurements, allowing the model to generate predictions without retraining for every new dataset.
This could be useful when investigating changes associated with infection, vaccination or other biological conditions. The model may also help identify missing experimental information and suggest which laboratory investigations could provide the most useful evidence.
Google DeepMind has made AlphaGenome 2 available for non-commercial research through its AlphaGenome API. Commercial researchers can access it through a private preview programme.
What AlphaGenome 2 Means for Future Disease Research
The introduction of AlphaGenome 2 reflects a broader effort to combine artificial intelligence with large-scale biological datasets. By connecting genetic variations with specific cell types, researchers may gain a clearer understanding of how certain DNA changes contribute to complex diseases.
However, computational predictions are not the same as experimentally confirmed biological mechanisms. The model is intended for research purposes and has not been validated or approved for clinical or diagnostic use.
Its practical value will depend on how researchers test its predictions and incorporate them into experimental studies. Nevertheless, AlphaGenome 2 offers a more detailed approach to studying genetic variation, potentially helping scientists identify promising research directions and understand disease biology at the cellular level.
