Skip to main content
The Quantum Dispatch
Back to Home
Cover illustration for AlphaGenome 2: What Cell-Level DNA Predictions Unlock

AlphaGenome 2: What Cell-Level DNA Predictions Unlock

AlphaGenome 2 predicts how DNA variants act in specific cell types, finding nearly twice as many causal variants as the first model. Why it matters.

Dr. Nova Chen
Dr. Nova Chen★Nov 13, 2026★4 min read

Google DeepMind has released AlphaGenome 2, a next-generation DNA sequence model that predicts how genetic variants change gene activity inside specific human cell types rather than across whole-tissue averages. Announced on November 11, 2026, it is available today for non-commercial research through the AlphaGenome API, with a private preview for commercial research.

  • What it predicts: the effect of DNA sequences and variants on one chosen tissue, cell type or cell state at a time, from a 1-million-letter input sequence
  • Training data: bulk and single-cell data from human and mouse samples, including over 107 million individual cell profiles
  • Headline result: on the TenK10K immune-cell dataset it identified 30% of causal variants versus 16% for the original model, and 43% of directional gene effects versus 23%
  • Access: AlphaGenome API for non-commercial research now; AlphaGenome 2 Skills in Google Antigravity are coming soon

What Is New in AlphaGenome 2?

The original AlphaGenome predicted thousands of experimental readouts in parallel across many tissues. According to DeepMind, that design spread the model's capacity thin, making it hard to separate closely related cell types. AlphaGenome 2 is a conditional model instead: alongside the DNA sequence, it takes a target biological condition as input at every stage of its architecture. That lets it learn how the same stretch of DNA is read differently in, say, a heart muscle cell versus an immune cell.

It still predicts the familiar readouts, such as gene expression, histone modifications and chromatin accessibility, and adds DNA methylation, which DeepMind describes as a widely requested output. Because the cell state is separated from the outputs, a scientist who has measured only one property can ask the model to impute others, which helps decide which experiment to run next.

How Does AlphaGenome 2 Handle Cells It Has Never Seen?

This is the part I find most promising. Researchers can upload a single-cell RNA-seq snapshot from just a few dozen cells, and the model compresses it into a cell-state representation that conditions its predictions without any retraining. DeepMind reports that this approach matches or beats models trained directly on those datasets. In one example using unseen data from a COVID-19 mRNA vaccine study, AlphaGenome 2 separated 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.

Why Cell-Specific Variant Prediction Matters for Disease Research

Genome-wide association studies flag broad regions of DNA linked to traits and diseases, but pinning down the exact causal variant, its target gene and the cell where it acts is slow work. DeepMind says AlphaGenome 2 computationally reproduced a recent finding that a distant regulatory variant drives inflammatory bowel disease by boosting the ETS2 gene in gut macrophages, while leaving other cell types largely untouched. The team has also precomputed predictions for trait-associated variants across hundreds of human cell and tissue conditions, so researchers can explore them without running large computations themselves.

That builds directly on last year's model and on AlphaGenome Atlas, which mapped all 9 billion possible single-letter DNA variants. The step from tissue averages to individual cell states is exactly what single-cell biology has been waiting for.

What Comes Next

DeepMind says it is working with groups including the Wellcome Sanger Institute to profile a wider range of cell types and developmental lineages, and plans to bring AlphaGenome 2 into Antigravity so agentic science tools can call it directly. For readers following AI in biology, our AI coverage tracks how specialized science models like this one keep moving from papers into tools researchers can actually use.

Sources: Google DeepMind: AlphaGenome 2 — November 11, 2026.

More AI Stories