AlphaGenome Atlas

AlphaGenome Atlas holds predictions for roughly 9 billion single-nucleotide variants, covering every single-letter change possible in the human genome. That scope is what separates it from querying a model on demand: the heavy computation is already done, so you get an answer instantly instead of waiting on a fresh inference run. You search, you get a score. No waiting.

The AVI score turns a pile of model outputs into a single, comparable number. It combines predictions from both AlphaGenome (which reads non-coding effects) and AlphaMissense (which covers protein-altering changes), so coding and non-coding variants land on the same scale. For teams triaging thousands of candidates, that one number is the difference between a usable shortlist and a spreadsheet nobody can sort. That's the whole point.

Most genetic tools focus on the protein-coding 2% of the genome because that part is easier to interpret. AlphaGenome Atlas also scores the other 98%, the non-coding regions that regulate how genes get switched on and off. If your variants fall outside known genes, this is exactly the territory that tends to go unexplored. Most tools skip it.

The Atlas ships as an intuitive website portal that requires zero coding skills. It works as a genetic research tool right in the browser. Clinical researchers and biologists can search, filter, and read results directly in the browser. That matters because a lot of the people who need these insights aren't software engineers, and forcing them into a scripting environment would exclude most of the audience.

The same predictions are reachable through the AlphaGenome API, the web portal, and a skill inside Google Antigravity. Researchers who want to automate workflows can pull results programmatically, while those who prefer point-and-click never touch code. One dataset, several entry points, depending on how you work. Pick the one that fits.

Behind the portal sits a 1-petabyte dataset, more than 30 times larger than the AlphaFold Database. The size is a direct consequence of precomputing every variant rather than computing on demand. For users, the practical payoff is speed and predictability: no waiting on server-side runs, no version drift between queries. Simple as that.

The Atlas isn't a demo. External collaborators have already applied it to unsolved rare disease cases and large cohort studies, including work on data from more than 54,000 UK Biobank participants. Seeing real research outcomes attached to a tool helps you judge whether it fits your own workflow before committing time to it. Does that matter for your project? It usually does.