AlphaGenome 2: Deciphering how cells read DNA to better understand disease

2026-11-17 · Google DeepMind

Overview

AlphaGenome 2 is a next-generation DNA sequence model announced on November 11, 2026, in Science. It predicts how DNA sequences and their variants affect specific human cell types and states, helping researchers uncover cellular drivers of disease. The work is by Adam Gayoso, Natasha Latysheva, Pushmeet Kohli and Žiga Avsec.

Why cell context matters

Every cell carries the same DNA blueprint, but different cell types and states switch different genes on or off. Efforts such as the Human Cell Atlas map gene expression across millions of cells, yet predicting how subtle DNA variants alter expression in cells remains an unsolved challenge and a key driver of traits and disease.

From bulk tissues to conditional cell-type modeling

AlphaGenome 1 predicted molecular activity across broad tissues from a one-million-letter DNA sequence and produced fixed parallel outputs for bulk tissues and cell lines. AlphaGenome Atlas extended this to predictions for all 9 billion possible single-letter changes in the human genome. However, the parallel approach spread model capacity thin and made subtle differences between closely related cell types hard to capture.

AlphaGenome 2 is a conditional model. Alongside a one-million-letter DNA sequence, it takes a biological condition as input at every stage of its architecture. The condition can be a tissue, cell type or state, chosen from a reference catalog or generated from a scientist's own single-cell experiment through a cell encoder. This lets the model learn how the same DNA stretch is read differently across cell types.

Training and outputs

The model was trained on bulk and single-cell data from human and mouse samples, including more than 107 million individual cell profiles across 68 tissues. It predicts core experimental readouts of DNA function such as gene expression, histone modifications and chromatin accessibility, and also DNA methylation. Its conditional design enables cross-modality imputation: by separating a cell's biological state from its outputs, it can take one measurement, such as single-cell gene activity, and predict unmeasured properties such as chromatin accessibility.

Performance

AlphaGenome 2 identifies nearly twice as many variants that change gene expression in cells compared with the previous state-of-the-art model. It improves cell-type specificity: on genome regions outside its training data, it explains significantly more of how gene expression differs across cell types within a tissue, including continuous changes as human blood stem cells mature into red blood cells and immune lineages. On the TenK10K dataset, a large single-cell map of 28 primary immune cell types, it identified nearly twice as many causal variants (30% vs. 16%) and directional gene effects (43% vs. 23%) as the original AlphaGenome fine-tuned on single-cell data and other specialized tools.

According to Professor Sarah Teichmann, co-founder of the Human Cell Atlas consortium and Professor of Stem Cell Medicine at the University of Cambridge, AlphaGenome 2 adapts directly to new single-cell data without retraining and can predict how DNA variants act in newly profiled cell states from even a small sample of cells.

Availability

AlphaGenome 2 is available today for non-commercial research via the AlphaGenome API, and in private preview for commercial research via Google Cloud.

Source