Google announced AlphaGenome Atlas, an AI resource that predicts the consequences of all ~9 billion possible single-base variants in the human genome by running them through its AlphaGenome model. The system focuses on identifying functional elements within non-coding DNA, which makes up over 97% of the genome but remains poorly understood. The authors note its practical value will ultimately depend on whether biologists adopt it over alternatives derivable from its training data.
Background
Google's AlphaGenome model, built on the same deep learning architecture as AlphaFold, applies AI to genomic sequence analysis rather than protein structure prediction. The AlphaGenome Atlas extends this work into a comprehensive variant-effect prediction resource for the entire human genome.
- Source
- Ars Technica
- Published
- Sep 10, 2026 at 12:34 AM
- Score
- 7.0 / 10