On September 8, Google DeepMind introduced AlphaGenome Atlas. It contains precomputed estimates of molecular effects for approximately nine billion possible single-letter changes in human DNA.

The AVI score combines AlphaGenome and AlphaMissense outputs. The release is a large atlas, not the base model’s debut. The announcement distinguishes available non-commercial use from planned commercial cloud availability for the atlas.

We see potential in experiment planning: instead of choosing randomly, a laboratory can begin with variants predicted to have an effect. Candidate rankings still need to fit its particular research question.

A prediction is neither a laboratory measurement nor clinical confirmation. The developers explicitly state that the tool is not validated or approved for clinical use. Independent data and specific biological conditions remain essential tests.

Optimistic editorial estimate: initial exploratory projects could take 1–3 months. Moving from selected variants to treatment requires further research and clinical studies; a database release cannot establish that timeline.