Google DeepMind released AlphaGenome Atlas on September 8, giving researchers searchable AI predictions for roughly 9 billion possible single-letter changes in human DNA. The atlas is intended to rank variants for laboratory investigation, not to diagnose patients.
A single-letter variant replaces one of DNA's four chemical letters with another. Some variants change a protein, while others alter when a gene is active or how a cell processes its genetic instructions. The difficulty is deciding which of the millions found in human genomes are likely to matter.
DeepMind generated the atlas by comparing AlphaGenome's predictions for reference and altered DNA sequences. A second model combines those changes with protein-impact estimates from AlphaMissense and evidence about which DNA positions have been conserved across species. The resulting score ranks a variant while preserving the component predictions researchers may want to inspect.
The accompanying technical report describes one previously unresolved case involving DNM1, a gene associated with severe neurological disease. AlphaGenome predicted that the variant would disrupt the processing of genetic instructions in brain cells. Experiments with modified gene fragments in five cell lines supported that mechanism, and the researchers recommended classifying the variant as likely disease-causing when the result was considered with existing clinical evidence.
One supported case does not validate all 9 billion predictions. The atlas is available through a web portal and programming interface for non-commercial research, and Google's terms prohibit using it for clinical decisions. Its value will depend on independent laboratories testing high-ranked variants across genes, cell types and patient populations.
Sources
- Cheng and colleagues, AlphaGenome Atlas technical report, September 2026; methods and rare-disease analysis, pages 2–3 and 7–9.
- Google DeepMind's September 8 announcement.
- AlphaGenome API: documentation, Atlas access and usage restrictions.
- Illustrative DNA image: PublicDomainPictures, Wikimedia Commons, CC0 1.0. This is a general DNA illustration.
