Google DeepMind unveils AlphaGenome 2 for cell-specific DNA analysis

Google DeepMind has announced AlphaGenome 2, a DNA sequence model designed to predict how genetic variants affect specific human cell types and states. The company says the system is intended to help researchers find disease-relevant changes that broad tissue averages can miss. ¹

The update matters because the same DNA can behave differently in different cells. A heart muscle cell, for example, switches on a different set of genes from an immune cell. AlphaGenome 2 is designed to take that biological condition into account while analysing sequences up to one million letters long, rather than treating every tissue as a single blended signal. ¹

What AlphaGenome 2 does

Google DeepMind says the model can predict gene expression, histone modifications, chromatin accessibility and DNA methylation. It can also use a target cell type or state as part of the input, allowing researchers to examine how the same sequence may be read differently across specialised cells. The company says the model can make predictions for newly profiled cell states without retraining when researchers provide a small sample of single-cell data. ¹

The model was trained on more than 100 million individual cell profiles from 68 tissues. The announcement gives a broader figure of more than 107 million profiles from human and mouse samples, reflecting the bulk and single-cell data used in training. Google DeepMind says AlphaGenome 2 identifies nearly twice as many variants that change gene expression in cells compared with its previous state-of-the-art model. ¹

Those figures are company-reported performance claims, not independent confirmation. The announcement links to a technical paper for the underlying methods and comparisons, but the model’s real value will depend on how well outside researchers reproduce the results across different datasets and disease questions. ²

Research access, not a clinical tool

AlphaGenome 2 is available for non-commercial research through the AlphaGenome API, while commercial research access is in private preview. Google DeepMind also says it plans to release AlphaGenome 2 skills in Google Antigravity, its agentic development platform. ¹

The company explicitly says AlphaGenome and AlphaGenome 2 are intended for research use only and have not been validated or approved for clinical or diagnostic purposes. That distinction matters: predicting a possible regulatory effect is not the same as proving that a variant causes disease in a patient. The model may help scientists prioritise experiments, but it does not replace laboratory evidence or medical assessment. ¹

Our opinion

AlphaGenome 2 is a meaningful shift in emphasis from broad tissue averages towards the specialised cells where disease biology often becomes visible. The announcement’s strongest promise is not that an AI model has solved genetic interpretation, but that it could make the next round of experiments more targeted. The sensible test is whether independent researchers can turn those predictions into repeatable biological discoveries rather than simply accepting a larger benchmark number.