Google DeepMind launched AlphaGenome Atlas, an AI powered database that maps the predicted functional impact of all 9 billion possible single letter DNA changes, allowing researchers to query their 1-petabyte data resource to rank variants and see how they may disrupt gene regulation, splicing, or protein function.
Launched on September 8, the AlphaGenome Atlas is an “integrated data resource that is roughly 30 times the size of AlphaFold database, and is used to predict the functional impact of all 9 billion possible single nucleotide variants (SNVs) across the human genome.
We’re launching AlphaGenome Atlas: an AI-powered searchable database mapping the predicted impact of all 9 billion possible single-letter DNA changes.
Here’s how it could help researchers better understand our biology 🧵 pic.twitter.com/SABFZW5SiR
— Google DeepMind (@GoogleDeepMind) September 8, 2026
“We used the AlphaGenome AI model to pre-calculate the regulatory impact of all 9 billion single-letter genetic changes, resulting in a massive, 1-petabyte dataset. Our new Atlas helps scientists rapidly query this vast information,” read an excerpt from a blogpost by Google.
According to Google, the Atlas also introduces AlphaGenome Variant Impact (AVI) score, the combined output of AlphaGenome, a 2025 sequence model and AlphaMissense, that ranks a variant from lower to higher predicted effect and can be broken down into contributing mechanisms.
“This single, easy-to-use score combines predictions for both coding and non-coding regions, allowing researchers to quickly prioritize the most promising avenues for research without sifting through thousands of data points,” said Google.
According to Google, the AlphaGenome Atlas is involved in accelerating research in areas like rare genomic variations and identifying rare, non-coding variants linked to complex traits.
Access to researchers has been granted through a no-code website, the AlphaGenome API, and a skill in Google Antigravity. Google said non-commercial use is available now while commercial access is planned on Google Cloud. The underlying model remains available separately for researchers who need custom predictions rather than the precomputed map.
