AlphaGenome by Google DeepMind: AI model predicts the effects of DNA mutations

Google DeepMind unveiled AlphaGenome on June 25, 2025 — an AI model that, from a DNA sequence up to 1 million base pairs long, predicts thousands of molecular effects of single-nucleotide variants and mutations in regulatory regions of the genome. In January 2026, the research was published in the journal Nature, and the model itself became available via a separate link.

What the model predicts

AlphaGenome takes as input a long DNA sequence — up to 1 million base pairs — and outputs predictions across thousands of molecular characteristics of its regulatory activity. To assess the effect of a specific variant, the model compares predictions for the mutated and non-mutated versions of the same sequence.

• Input — a DNA sequence up to 1,000,000 base pairs long
• Training data was collected by the ENCODE, GTEx, 4D Nucleome, and FANTOM5 consortia
• The model covers hundreds of human and mouse cell and tissue types
• It predicts gene start and end sites, splice sites, the amount of RNA produced, DNA accessibility, and its binding to proteins
• The model's authors are Ziga Avsec and Natasha Latysheva of Google DeepMind

How AlphaGenome differs from AlphaMissense

AlphaGenome closes a gap that DeepMind's earlier model, AlphaMissense, didn't cover: that model specializes in variants within protein-coding regions, which make up just 2% of the human genome. The remaining 98% of the genome — non-coding regions — controls gene activity and contains many disease-associated variants, but it is precisely these that had resisted accurate prediction of mutation effects.

AlphaGenome also builds on DeepMind's earlier genomics model, Enformer, adding processing of longer sequences and predictions at higher resolution. The architecture combines convolutional layers to detect short patterns in DNA, transformers to exchange information across all positions in the sequence, and final layers that turn the detected patterns into predictions across different molecular modalities. During training, computation for a single sequence is distributed across several interconnected tensor processing units (TPUs).

How to access the model

Google DeepMind has opened access to AlphaGenome through its own API in preview mode — for now only for non-commercial research, with plans to release the model more broadly in the future. According to the company's blog, in January 2026 the work on AlphaGenome was published in the journal Nature, after which the full paper and access to the model became available via separate links.

"We believe

AlphaGenome can become a valuable resource for the scientific community and will help scientists better understand how the genome works, the biology of disease, and ultimately make new discoveries and develop new treatments," Google DeepMind's blog states.

What this means

AlphaGenome gives researchers a single tool for quickly assessing the effect of a specific mutation on the genome's regulatory activity, instead of a set of narrowly specialized models for each individual task. For disease genetics, this means faster initial screening of disease-associated variants before costly lab experiments.

Frequently asked questions

Who has access to AlphaGenome right now?

As of the June 25, 2025 announcement, access is available through the API in preview mode only for non-commercial research purposes; Google DeepMind plans a broader release of the model later.

How does AlphaGenome differ from AlphaMissense?

AlphaMissense evaluates variants in protein-coding regions of the genome, which make up about 2% of all DNA, while AlphaGenome covers non-coding regulatory regions, spanning the remaining 98% of the genome and many disease-associated variants.