AlphaGenome Atlas publishes predictions for nine billion DNA variants

Google DeepMind introduced AlphaGenome Atlas on 8 September, providing predicted effects for nine billion single-letter DNA variants. The resource precomputes AlphaGenome outputs and exposes them through a free academic-research portal, API and Antigravity skill. It also adds an AlphaGenome Variant Impact score combining signals from AlphaGenome and AlphaMissense.

The distinction between prediction and experiment is essential. A catalogue at that scale can help researchers prioritise variants, but it does not mean every listed effect has been measured in a laboratory or linked conclusively to disease. DeepMind described collaborators experimentally checking selected findings, rather than validating the entire atlas. The release is therefore research infrastructure for narrowing questions, not a diagnosis engine or a substitute for clinical genetic interpretation.