Google DeepMind releases an AI atlas mapping nine billion human DNA variants for researchers.

Beyond the roughly 9 billion single-nucleotide variants, the technical release also includes predictions for more than 100 million insertions and deletions observed in the gnomAD, UK Biobank and All of Us datasets.
Each variant is associated with an average of about 27,000 experiment-specific scalar predictions covering hundreds of human and mouse cell types and tissues, providing substantially more granular data than a single overall impact estimate.
The AlphaGenome Variant Impact score combines AlphaGenome and AlphaMissense predictions with evolutionary conservation and protein loss-of-function features; its accompanying feature attributions indicate which processes, such as gene expression or RNA splicing, drive a variant’s score.
The Atlas also contains a catalogue of 2,601 recurrent DNA sequence motifs—regulatory patterns sometimes described as the genome’s “words”—with their locations mapped across cell types.
In addition to its website and API, the resource is offered as a skill within Google’s agentic platform, Antigravity, giving researchers another way to query the precomputed predictions.
Google DeepMind has launched AlphaGenome Atlas, an AI tool that maps the biological effects of roughly 9 billion possible DNA mutations across the human genome. According to News Medical, the searchable resource predicts how each single-letter DNA change affects gene expression, RNA splicing, and other molecular processes without requiring researchers to have specialized computing skills or access to expensive equipment.
The 1-petabyte dataset includes predictions for more than 100 million insertions and deletions from major genetic databases. Google DeepMind offers free access to academic researchers, with potential applications spanning disease research, drug discovery, and personalized medicine—though computational predictions still need laboratory validation before clinical use.}
Each of the 9 billion variants includes an average of about 27,000 experiment-specific predictions covering hundreds of human and mouse cell types and tissues. Gigazine reports that the AlphaGenome Variant Impact score combines multiple data sources—including evolutionary conservation and protein loss-of-function features—to rank variants by importance in both coding and noncoding DNA regions.
The score includes feature attributions that show which molecular processes—such as gene expression or RNA splicing—drive each variant's impact. This granular approach gives researchers far more detailed information than older methods that provided only a single overall impact estimate.
Researchers can query the precomputed predictions through a web browser, an application programming interface (API), or as a skill within Google's agentic platform called Antigravity. Health and Me notes that the free access to academic researchers removes the barrier of needing specialized computational resources or advanced coding knowledge to search billions of variant predictions.
The atlas also catalogs 2,601 recurrent DNA sequence motifs—regulatory patterns sometimes described as the genome's "words"—mapped across different cell types. This regulatory mapping helps researchers understand how mutations affect not just individual genes but broader genetic control systems.
DeepMind emphasizes that these predictions are computational estimates, not clinical proof. Dainik Jagran reports that researchers must conduct laboratory validation and address privacy, regulatory, and ethical concerns before applying the atlas results to medical workflows or patient care decisions.
The atlas accelerates the discovery phase by helping researchers prioritize which variants and DNA regions to study in the lab. By narrowing down billions of possibilities to the most promising candidates, the tool saves time and resources while maintaining the rigorous scientific standards required for medical applications.
Publishers
24
Articles
12
Reach
36