Google DeepMind Launches AlphaGenome 2 for Cell-Specific DNA

Google DeepMind AlphaGenome 2 launch

Google DeepMind has unveiled AlphaGenome 2, a next-generation artificial intelligence model built to predict how DNA sequences and their variations affect specific human cell types. The announcement dropped on November 11, 2026, from a team led by Adam Gayoso, Natasha Latysheva, Pushmeet Kohli, and Žiga Avsec. The company says the system bridges single-cell biology and human genetics, giving researchers a fresh lens for uncovering the cellular drivers of disease.

Every cell in the human body carries the same genetic blueprint, yet heart muscle cells, immune cells, and neurons each switch different genes on or off to do their specialized jobs. Large-scale efforts like the Human Cell Atlas have mapped gene expression across millions of cells, but predicting how subtle DNA changes alter that expression in specific cell types has remained a stubborn nut to crack. Those changes are a key driver of human traits and disease susceptibility.

AlphaGenome 1 vs AlphaGenome 2 differences

Last year, DeepMind released the first AlphaGenome model to predict how DNA variants affect molecular activity across broad tissues. The team later expanded its reach with AlphaGenome Atlas, a platform holding predictions for all nine billion possible single-letter changes in the human genome. Those tools have already helped researchers prioritize and interpret variant effects, driving discoveries in rare diseases, complex traits, and fundamental biology. AlphaGenome 2 represents a step-change improvement by zooming in from tissue-level averages to individual cell types and states.

The original AlphaGenome took a one-million-letter DNA sequence and predicted properties across thousands of experiments from many tissues all at once. That parallel approach spread the model's capacity thin and made it tough to capture subtle differences between closely related cell types. DeepMind reports it fundamentally redesigned how cell types are treated. AlphaGenome 2 is a conditional model that predicts DNA effects for one biological condition-defined as a tissue, cell type, or state-at a time. Alongside the DNA sequence, the target condition is fed into every stage of the architecture, letting the model learn how the same stretch of DNA is read differently across cell types.

This conditional design also enables cross-modality imputation. The model separates a cell's biological state from its outputs, so if a scientist provides just one measurement like single-cell gene activity, AlphaGenome 2 can predict unmeasured properties like chromatin accessibility. That capability helps researchers decide which experiments to run next.

AlphaGenome 2 training data scale

DeepMind says it trained AlphaGenome 2 on both bulk and single-cell data from human and mouse samples, including more than 107 million individual cell profiles across 68 tissues. Moving beyond the bulk tissue averages of the original model, the new system can resolve activity inside specialized cells like distinct heart muscle or blood immune cells. The company claims the model identifies nearly twice as many variants that change gene expression in cells compared with its previous state-of-the-art model. It also says AlphaGenome 2 can predict variant effects in disease-relevant cells that would be missed in broad-tissue models, plus effects for cell and disease states the model has never seen before.

The model predicts the same core experimental readouts of DNA function as its predecessor, including gene expression, chromatin accessibility, and transcription factor binding. DeepMind is making AlphaGenome 2 available today for non-commercial research through the AlphaGenome API, with a private preview for commercial research.

Conditional AI architecture biology

For people who build with AI, the release signals a shift toward conditional architectures that can handle biological context as a first-class input rather than a fixed output channel. The cross-modality imputation feature also shows how a single model can replace multiple specialized predictors, potentially simplifying pipelines that currently stitch together separate tools for expression, accessibility, and binding data. Researchers who work with single-cell data should evaluate whether the API can accelerate variant prioritization in their own datasets, particularly for cell types that are underrepresented in bulk-tissue studies. Those planning commercial applications will need to engage with DeepMind's private preview program to understand licensing terms and data governance requirements.

Frequently asked questions

When did Google DeepMind release AlphaGenome 2 and who led the team?

Google DeepMind released AlphaGenome 2 on November 11, 2026. The team was led by Adam Gayoso, Natasha Latysheva, Pushmeet Kohli, and Žiga Avsec.

How does AlphaGenome 2 differ from the original AlphaGenome model?

The original AlphaGenome predicted DNA effects across broad tissue averages simultaneously. AlphaGenome 2 is a conditional model that predicts effects for one specific cell type or state at a time, with the target condition fed into every architecture stage, enabling it to capture subtle differences between closely related cell types.

What new capability does AlphaGenome 2's conditional design enable?

The conditional design enables cross-modality imputation. By separating a cell's biological state from its outputs, AlphaGenome 2 can predict unmeasured properties like chromatin accessibility when given just one measurement such as single-cell gene activity.

What training data was used for AlphaGenome 2?

AlphaGenome 2 was trained on both bulk and single-cell data from human and mouse samples, including more than 107 million individual cell profiles across 68 tissues.

How can researchers access AlphaGenome 2?

AlphaGenome 2 is available today for non-commercial research through the AlphaGenome API. Commercial researchers can apply for a private preview program to understand licensing terms and data governance requirements.

Source: Google DeepMind