# AlphaGenome 2 — DeepMind's cell-type-aware DNA effect model

> A DNA model that predicts variant effects for one chosen cell type at a time, and adapts to new single-cell data.

- Org: Google DeepMind
- Date: 2026-11-11
- Category: model, paper
- Source: https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/
- Canonical page: https://ai-tldr.dev/releases/google-alphagenome-2/

## In short

AlphaGenome 2 is a Google DeepMind model that predicts how DNA variants change gene activity in a chosen human cell type. It trained on 100M+ single-cell profiles across 68 tissues and is free for non-commercial research.

## What it is

AlphaGenome 2 is the successor to DeepMind's AlphaGenome, released a year earlier. It reads a DNA sequence of about one million letters and predicts how that sequence, and any variant in it, changes gene expression, histone marks, chromatin accessibility and, new in this version, DNA methylation.

## How it works

The first model predicted many tissues in parallel. The new one is conditional: it takes the target tissue, cell type or state as an input and uses it at every stage. Conditions come from a reference catalog, or from a user's own single-cell experiment through a cell encoder. It can also impute unmeasured readouts from one measured signal.

## Why it matters

DeepMind says the model finds nearly twice as many variants that change gene expression as the previous best model. On the TenK10K immune-cell data it identified 30% of causal variants versus 16%, and 43% of directional gene effects versus 23%. That helps researchers link disease variants to the cell types where they act.

## Who it's for

Genomics and single-cell researchers

## Try it

Request access through the AlphaGenome API docs at alphagenomedocs.com

## Quick facts

- Maker: Google DeepMind
- Input: ~1M-letter DNA sequence + a target cell type or state
- Training data: 100M+ single-cell profiles, 68 tissues, human and mouse
- New readout: DNA methylation
- Availability: AlphaGenome API, non-commercial research
- Commercial use: Private preview

## FAQ

### Can AlphaGenome 2 be used commercially?

AlphaGenome 2 is available through the AlphaGenome API for non-commercial research. Google DeepMind offers a private preview for commercial research. The post does not state a separate software license, and the models are intended for research use only, not clinical or diagnostic use.

### How is AlphaGenome 2 different from the first AlphaGenome?

AlphaGenome 2 is a conditional model that predicts one tissue, cell type or state at a time, where the first AlphaGenome predicted many tissues in parallel. It adds DNA methylation as a readout and a cell encoder that accepts a user's single-cell data, so it can adapt without retraining.

### How well does AlphaGenome 2 find causal variants?

On the TenK10K dataset of 28 immune cell types, AlphaGenome 2 identified 30% of causal variants against 16% for AlphaGenome fine-tuned on single-cell data. It got directional gene effects right 43% of the time against 23%, according to Google DeepMind's announcement.

### Is AlphaGenome 2 approved for diagnosis?

No. Google DeepMind states that AlphaGenome 2 outputs are intended for research use only and have not been validated or approved for clinical or diagnostic purposes.

## Links

- [Announcement](https://deepmind.google/blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/)
- [Paper (PDF)](https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphagenome-2-deciphering-how-cells-read-dna-to-better-understand-disease/alphagenome-2.pdf)
- [API Docs](https://www.alphagenomedocs.com/)
- [AlphaGenome Launch](https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/)