On November 11, Google DeepMind released AlphaGenome 2, a model that predicts what a DNA variant does inside a specific human cell type. Instead of giving tissue-level averages like its predecessor, it answers for one cell type or cell state at a time. The model is available from today for non-commercial research through the AlphaGenome API, with commercial research in private preview.
The change comes from how it was trained. DeepMind trained it on more than 107 million individual human and mouse cell profiles across 68 tissues. The model takes a DNA sequence of up to one million letters plus a chosen tissue, cell type or cell state, and predicts gene expression, chromatin accessibility and related outputs for that condition. Researchers can also condition it on their own single-cell experiment, using as few as a few dozen cells, without retraining. On the TenK10K immune-cell benchmark, it recovered 30% of putatively causal expression-linked variants at 90% precision, against 16% for the first model fine-tuned on single-cell data; accuracy on gene effect direction rose from 23% to 43%.
The same variant behaves differently in different cells
Much disease genetics has been stuck on a blind spot: a variant may matter only in one cell type, and its signal gets diluted in a tissue average. By predicting at cell level, AlphaGenome 2 lets researchers pin down which cell type drives a disease before choosing the next experiment. Genomics AI has more than one route: DeepGEM 2.0 pre-screens lung cancer gene targets from pathology slides, which goes from images to genes. For general readers, the direct meaning is not medical: DeepMind states clearly that AlphaGenome 2 is for research only, has not been validated for clinical or diagnostic use, and must not be used to diagnose or guide treatment. Non-commercial research access is free, commercial access waits on private preview, and the model weights are not public.
Three things are worth watching next: how many new disease associations researchers find with it, when the promised Antigravity skills arrive, and which teams get extra quota once the community spotlight program closes on November 30. AlphaGenome 2 does not replace its predecessor — the older model keeps serving genome-wide atlases, while the new one handles finer, cell-level questions.
