Ordelvio News

Science · Published

Google DeepMind unveils AlphaGenome 2 to predict DNA variant effects in specific cells

The company said its new model finds nearly twice as many variants that change gene expression in cells as its previous model.

Illustration in a hand-printed ink style in coral ink on paper white.AI illustration
AI illustration · not a photograph of the event · How AI is used

Google DeepMind has announced AlphaGenome 2, a model that predicts how DNA sequences and their variants affect gene activity in specific human cell types. The company said the model identifies nearly twice as many variants that change gene expression in cells as its previous model.

The first AlphaGenome, launched last year, predicted how DNA variants affect molecular activity across broad tissues. According to Google DeepMind, it predicted properties for thousands of experiments at once, which spread its capacity thin and made subtle differences between closely related cell types hard to capture.

AlphaGenome 2 instead works as a conditional model. It takes a million-letter DNA sequence together with one target condition, such as a tissue, cell type or cell state, and makes predictions for that condition alone. The company said this lets the model learn how the same piece of DNA is read differently in different cells. It was trained on more than 100 million individual cell profiles across 68 tissues, using human and mouse data. Besides gene expression, histone modifications and chromatin accessibility, it now also predicts DNA methylation.

On the TenK10K dataset, which covers 28 primary immune cell types, the model identified 30% of causal variants against 16% for the original AlphaGenome fine-tuned on single-cell data and other specialized tools, and 43% of directional gene effects against 23%, the company said. It also reproduced, computationally, an earlier study's finding that a variant drives IBD by overactivating the ETS2 gene in gut macrophages.

Google DeepMind said the model can be applied to new single-cell experiments without retraining, using data from just a few dozen cells. In data from a COVID-19 mRNA vaccine study it had not seen, the model separated vaccine-activated T cells from resting ones and showed that a variant linked to severe COVID-19 disrupts gene activity more strongly in the activated state.

The company has also precomputed predictions for trait-associated variants across hundreds of human cell and tissue conditions. AlphaGenome 2 is available for non-commercial research through the AlphaGenome API and in private preview for commercial research. Google DeepMind said the models are intended for research use only and have not been validated or approved for clinical or diagnostic purposes.

  • Genomics
  • DNA
  • AI Models
  • Single-Cell Biology
  • Google DeepMind