There are roughly 9 billion ways a single DNA letter can change in the human genome. Testing each one in a laboratory is essentially impossible. So Google DeepMind did the next best thing: it computed all of them.

AlphaGenome Atlas is a new platform containing precomputed predictions for every possible single-nucleotide variant in the human genome. That’s 9 billion variants, each annotated with thousands of molecular effect predictions across hundreds of human and mouse cell types and tissues. The result is a 1-petabyte dataset — more than 30 times larger than the AlphaFold Database — and it’s freely accessible to academic researchers through a web portal that requires no coding experience.

What AlphaGenome Atlas actually does

The platform is built on AlphaGenome, an AI model Google DeepMind released earlier to predict how genetic variants affect biological processes. AlphaGenome is useful for analyzing specific variants, but running it on individual queries limits how broadly researchers can apply it. By precomputing its predictions at genome-wide scale, the Atlas gives scientists instant access to results that would otherwise take enormous time and compute resources to generate.

Think of it the way the company describes it: an atlas is a collection of maps, each charting different features of the same terrain. AlphaGenome Atlas does the same for the genome, linking variant locations to their predicted molecular effects across gene regulation, chromatin accessibility, RNA splicing, and protein function.

Alongside the Atlas, Google DeepMind is releasing the AlphaGenome Variant Impact score, or AVI. It condenses predictions from both AlphaGenome and AlphaMissense — the company’s earlier model for protein-altering variants — into a single number per variant. Researchers can use it to rapidly rank variants by predicted impact and then trace which biological processes are most disrupted. That combination of speed and interpretability is exactly what’s been missing in variant prioritization workflows.

What’s inside the platform

The Atlas offers several interconnected data layers:

• Molecular effect predictions across thousands of gene regulatory outputs, spanning hundreds of cell types and tissues in both human and mouse
• AVI scores — a single impact number for each of the 9 billion variants, covering both coding and non-coding regions of the genome
• AVI feature attributions that break down each score into contributing biological mechanisms, such as splicing changes or shifts in gene expression
• A compendium of more than 2,500 recurring DNA sequence motifs, effectively a catalogue of the genome’s functional vocabulary

The non-coding coverage is worth highlighting. Only about 2% of the genome codes for proteins — the part genetics has historically focused on. The remaining 98% regulates when and how genes are switched on or off, and it’s where the majority of trait-associated variants actually sit. Most existing variant scoring tools struggle here. Google DeepMind says the AVI score performs well across both regions, and independent benchmarking against rare disease and variant pathogenicity datasets supports that claim.

Early results in rare disease research

Academic collaborators have already applied the Atlas to real research problems. Working with the GREGoR Consortium, researchers at the Broad Institute used AVI scores to prioritize variants in unsolved rare disease cases. In one instance, the approach surfaced a previously overlooked variant in a gene called DNM1, which is associated with epileptic encephalopathy. The AlphaGenome predictions didn’t just flag the variant — they showed the mechanism: the mutation created a false splice site, causing an abnormal extension of the resulting protein. Experimental validation confirmed the finding and identified nearby variants with similar effects.

That kind of mechanistic specificity matters. Knowing a variant is likely pathogenic is useful. Knowing why, at the molecular level, is what enables follow-on research and, eventually, therapeutic development.

Availability and access

AlphaGenome Atlas is available now through a free web portal, the AlphaGenome API, and as a skill within Google Antigravity. The web interface is designed for researchers without bioinformatics expertise, following the same accessibility-first approach Google DeepMind used when it opened the AlphaFold Database in 2022. That database, which grew from around 190,000 experimental protein structures to more than 200 million predictions, became a standard resource across the life sciences within months of launch.

Whether AlphaGenome Atlas achieves comparable adoption will depend on how well the predictions hold up across diverse research applications. But the scale is real, the early validation is promising, and the barrier to access is low. For researchers working on rare diseases, population genetics, or the functional biology of non-coding DNA, this is worth exploring now.