Google has released AlphaGenome Atlas, a resource that uses its AlphaGenome AI system to predict the functional consequences of every possible single-base change across the human genome. The tool focuses on non-coding DNA, which makes up the vast majority of the genome but does not encode proteins; some of this DNA regulates gene activity, while much of it is thought to be evolutionary debris from ancient viruses and genetic parasites. Distinguishing which sequences matter has long been difficult for biologists, since DNA-binding proteins are not highly selective and differ between cell types, making AI-based probability modelling a promising approach.

The human genome contains roughly 3 billion bases, and testing each position against the three alternative DNA letters not present in the reference sequence meant running around 9 billion base substitutions through the AlphaGenome software. The system predicts effects such as gene expression, chromatin accessibility, histone modification, transcription factor binding and splicing, and has so far been trained only on mouse and human sequences using a limited set of well-studied cell types. Its predictions reportedly match or exceed those of existing specialised tools, offering researchers a quick way to gauge whether a genetic variant is likely to be significant, though its real-world value will only become clear once biologists adopt it more widely.