AlphaGenome Atlas maps every possible DNA mutation in the human genome

Nine billion. That is the number of possible single-letter changes in the human genome, and until now, researchers had no practical way to assess them all. Google DeepMind has changed that with the launch of AlphaGenome Atlas, a free platform that pre-computes predictions for every one of those variants and makes the results available to the global research community through a web portal requiring no coding experience.

The scale alone is striking. AlphaGenome Atlas is a 1-petabyte dataset, more than 30 times the size of the AlphaFold Database, which itself was considered a landmark achievement when it expanded to cover more than 200 million protein structure predictions in 2022. That database became a standard reference tool across the life sciences. DeepMind is clearly aiming for the same outcome here.

The platform builds on AlphaGenome, an AI model DeepMind released earlier that predicts how specific genetic variants affect biological processes. Useful for targeted analysis, AlphaGenome was nonetheless limited in scope. Atlas changes the equation by running those predictions at genome-wide scale and storing the results, so researchers can access findings instantly rather than waiting for individual computations.

How does it work?

At its core, AlphaGenome Atlas contains thousands of molecular effect predictions for each of the 9 billion single-nucleotide variants in the human genome. These predictions cover multiple aspects of gene regulation and span hundreds of human and mouse cell types and tissues. But the platform is more than a raw data repository. It also includes several interconnected tools:

• Molecular effect predictions: Thousands of predictions per variant, covering gene regulation across hundreds of cell types and tissues in both humans and mice.
• AlphaGenome Variant Impact (AVI) score: A single numerical score summarising the predicted impact of each variant, combining outputs from AlphaGenome and AlphaMissense, DeepMind's earlier model for protein-altering variants.
• AVI feature attributions: Explanations linked to each AVI score, identifying which biological processes, such as RNA splicing or gene expression, are predicted to be most affected.
• DNA sequence motifs: A catalogue of more than 2,500 recurring DNA sequences and their locations across the genome, essentially a reference library of the genome's functional "words".

The AVI score is particularly significant because it works across both coding and non-coding regions of the genome. Coding regions make up roughly 2% of the human genome. The remaining 98%, the non-coding portion, regulates gene activity and contains the majority of variants associated with common traits and disease. Most existing tools perform poorly in this space. DeepMind says its internal testing shows the AVI score achieves best-in-class performance across multiple variant pathogenicity and rare disease benchmarks.

Why does it matter?

For researchers working on rare diseases, the practical value is already being demonstrated. In collaboration with the GREGoR Consortium, scientists at the Broad Institute used the AVI score to prioritise variants in cases where previous analysis had come up short. In one example, the team identified a variant affecting a gene called DNM1, strongly associated with epileptic encephalopathy. The AlphaGenome predictions did not just flag the variant as significant; they explained the mechanism. The variant was creating an incorrect splice site, causing an abnormal extension of the resulting protein. Experimental validation confirmed the prediction and identified additional nearby variants with similar effects.

That combination of prediction and mechanistic explanation is what separates Atlas from a simple filtering tool. Researchers are not just getting a ranked list. They are getting a biological rationale they can test in the lab.

And the applications extend beyond rare disease. Identifying which non-coding variants are linked to common traits in the general population has long been difficult because the volume of harmless genetic variation creates statistical noise that obscures meaningful signals. Atlas is designed to cut through that noise, giving population geneticists a sharper instrument for large-scale association studies.

The context

For health systems and research institutions across the GCC, this kind of tool has direct relevance. Several Gulf states carry elevated rates of certain genetic conditions, partly because of historically high rates of consanguinity in some communities, and partly because regional populations have been underrepresented in global genomic databases built largely on European ancestry data. Saudi Arabia's Vision 2030 health agenda and the UAE's broader digital health strategy both place genomics and precision medicine at the centre of future healthcare planning. Tools that make genome-wide variant analysis accessible without specialist bioinformatics infrastructure lower the barrier for regional institutions to participate in this research.

AlphaGenome Atlas is available now through a web portal, via the AlphaGenome API, and as a skill in Google Antigravity. Access is free for academic research. A companion paper has been published detailing the methodology and benchmark results.

The comparison to AlphaFold is worth taking seriously. That database did not just speed up protein structure research; it changed what questions scientists thought were worth asking. If AlphaGenome Atlas achieves even a fraction of that effect in genomics, the implications for rare disease diagnosis, drug target identification, and population health research across the region and beyond will be substantial.