Loading…
Loading…
AlphaGenome Atlas turns roughly nine billion possible single-letter DNA changes into a research map. What evidence should turn a ranked candidate into a decision?
This Frontier Brief explains the AlphaGenome Variant Impact (AVI) score, its use of AlphaMissense, and DeepMind's reported DNM1 research example, where researchers tested a predicted molecular effect experimentally.
The product question: keep the predicted effect, its uncertainty, and the independent evidence attached to the recommendation — with a clear owner for the next decision.
AlphaGenome is not validated or approved for clinical use. This discusses research and product design, not medical advice. The molecular diagrams and candidate ranking are conceptual illustrations, not patient data, measured scores, or an Atlas screen recording. The reported research findings are attributed to DeepMind and its collaborators.
Nine billion predictions is a map, not a set of answers — ranking is where the work starts, not where it ends.
The map predicts molecular effects of single-letter changes, not clinical outcomes.
The AVI score ranks variants by predicted impact, drawing on AlphaMissense.
The DNM1 example matters because a predicted molecular effect was tested experimentally, not because it ranked highly.
Access comes through three surfaces: the portal, the API, and Antigravity.
AlphaGenome ships under research-use terms and is not validated or approved for clinical use.
Keep the predicted effect, its uncertainty, and the independent evidence attached to the recommendation all the way to the decision.
Every ranked candidate needs a named owner for the next decision.
Nine billion predictions
What the map predicts
How AVI ranks variants
The AlphaFold distribution strategy
The DNM1 research example