Google DeepMind has unveiled AlphaGenome 2, a genomics AI model capable of predicting how DNA variants affect specific human cell types. Trained on over 107 million individual cell profiles across 68 human and mouse tissues, the new system moves past bulk tissue predictions to pinpoint cellular drivers of disease with significantly higher precision than its predecessor.

Unlike the original AlphaGenome, which evaluated static DNA sequences across broad tissues, AlphaGenome 2 uses a conditional architecture taking both DNA sequences and specific cell states as inputs. The model predicts gene expression, physical DNA packaging within the cell nucleus, and DNA methylation switches, while also inferring missing molecular data from single laboratory measurements.

In benchmark testing on 28 primary immune cell types, AlphaGenome 2 successfully identified 30 percent of causal variants, nearly doubling the performance of AlphaGenome 1. Furthermore, the architecture adapts to new experiments using data from just a few dozen cells without requiring full model retraining.

Why it matters

  • Enables biotech founders and researchers to accurately model cell-type-specific genetic variants without expensive, full-scale lab retraining.

  • Improves targeted drug discovery by predicting DNA packaging, methylation, and gene regulation at single-cell resolution.

  • Demonstrates zero-shot adaptation to novel cell states using minimal experimental data samples.

Source: the-decoder.com