ResearchNews
AlphaGenome 2 predicts DNA variant effects per cell type
Google DeepMind released AlphaGenome 2. Instead of tissue averages, the DNA model now predicts how genetic variants act in specific cell types and cell states, and it identifies nearly twice as many variants that change gene expression as its predecessor.
· 2 min read
What the new model does
The original AlphaGenome launched last year and predicted variant effects across broad tissues. AlphaGenome 2 zooms in to individual cell types, such as heart muscle cells or immune cells (Google DeepMind).
- Training data: more than 107 million individual cell profiles across 68 tissues, from human and mouse samples.
- Conditional model: alongside the million-letter DNA sequence it takes the target cell type or state as input, so it learns how the same DNA is read differently across cells.
- New readout: besides gene expression, histone modifications and chromatin accessibility, it now predicts DNA methylation.
- Cross-modality imputation: from one measurement, such as single-cell gene activity, it predicts unmeasured properties.
The numbers
On TenK10K, a single-cell map of 28 primary immune cell types, AlphaGenome 2 identified 30% of causal variants versus 16% for the original AlphaGenome fine-tuned on single-cell data. For directional gene effects the figure is 43% versus 23%.
It also reproduced a recent IBD finding computationally: it singled out the causal variant and showed that it boosts the ETS2 gene in gut macrophages while leaving other cell types largely unaffected.
New cell states without retraining
Researchers can upload a single-cell RNA-seq snapshot from just a few dozen cells. The model compresses it into a cell state representation and predicts for that state without retraining. On data from a COVID-19 mRNA vaccine study, it separated activated T cells from resting ones.
Who can use it
It is available now for non-commercial research through the AlphaGenome API, with Colab tutorials, and in private preview for commercial research. It is not approved for clinical diagnosis. Genomics CEO Peter Donnelly calls it a step change, but notes that it makes confident predictions for only a small subset of the variants that can be queried (Genomics).
For developers this is a scientific model behind an API, not a chatbot: something to plug into a bioinformatics pipeline for variant prioritization. For teams in drug discovery or rare disease research, the cell-type precision is the real news.
Produced with the help of AI, reviewed by an editor. How we make our articles
Reviewed and edited by the Neural Daily desk. About
