Google DeepMind's AlphaGenome 2 Identifies 30% of Causal Variants

Google DeepMind announced AlphaGenome 2 on Wednesday, Nov. 11. In the TenK10K map of 28 primary immune cell types, the model identified 30% of causal variants, compared with 16% for the original AlphaGenome fine-tuned on single-cell data. Its directional gene-effect result was 43%, versus 23% for the earlier model. The benchmark measures model performance, not clinical outcomes.

AlphaGenome 2 analyzes a DNA sequence of up to one million letters alongside a target such as a tissue, cell type or cell state. It predicts gene expression, histone modifications, chromatin accessibility and DNA methylation.

The original AlphaGenome made predictions across broader tissue averages. The newer model can use single-cell RNA-sequencing data to study new cell states without retraining. It was trained on bulk and single-cell data from human and mouse samples, totaling more than 107 million individual cell profiles.

“What makes AlphaGenome 2 exciting is how it adapts directly to new single-cell data without retraining,” said Sarah Teichmann, co-founder of the Human Cell Atlas consortium and professor of stem cell medicine at the University of Cambridge.

Noncommercial researchers can access AlphaGenome 2 through its API; commercial research access is in private preview. The model has not been validated or approved for clinical or diagnostic use.