DeepMind releases AlphaGenome Atlas covering 9 billion DNA mutations

September 14, 2026 at 12:27 AM2 min

This digest was compiled by AI from multiple sources — links to the originals are below.

Google DeepMind has released AlphaGenome Atlas, a database of predicted effects for all 9 billion possible single-letter mutations in human DNA. The atlas, built on the AlphaGenome model, includes over 100 million short insertions and is available free for non-commercial use. It aims to help researchers prioritize disease-causing variants, including in non-coding DNA.

Key Facts

AlphaGenome Atlas covers all 9 billion possible single-letter mutations in the human genome.
The database is approximately 1 petabyte in size and includes over 100 million short insertions.
DeepMind introduced a unified AlphaGenome Variant Impact (AVI) score to prioritize variants for further study.
The atlas is freely available for non-commercial use.
Predictions from the atlas helped Broad Institute identify a non-coding variant as a possible cause of severe epilepsy.

Atlas Coverage and Scale

AlphaGenome Atlas contains predictions for every possible single-letter substitution in the human genome, totaling 9 billion variants. The dataset is approximately 1 petabyte in size. It also includes more than 100 million short insertions and DNA fragments found in human genomes. For each mutation, the atlas provides thousands of predicted effects, including changes in nearby gene activity across tissues and alterations in chromatin structure.

Variant Impact Scoring

DeepMind added a unified AlphaGenome Variant Impact (AVI) score for each mutation to help researchers quickly identify variants worth further study. In tests, AVI distinguished disease-causing mutations from benign changes. The atlas is intended for variant search and prioritization, not for standalone diagnosis. It is particularly valuable for non-coding DNA, which makes up 98% of the human genome and remains poorly understood.

Research Applications and Limitations

Predictions from the atlas helped a Broad Institute team identify a variant in non-coding DNA as a possible cause of severe epilepsy. Using AlphaGenome predictions, DeepMind mapped thousands of short sequence motifs and proposed their roles in different cell types, including gene activation, repression, and DNA accessibility. The atlas lowers the computational barrier for geneticists by providing ready-made predictions instead of requiring direct model access. The system analyzes mutations one at a time relative to a reference genome and does not account for combinations of variants or individual genetic differences. Authors and independent experts emphasize that predictions do not replace experiments and should not be used alone for clinical decisions.