Google DeepMind has made AlphaGenome Atlas available, predicting in advance the biological effects of approximately 9 billion possible single-nucleotide variants in the human genome. The platform brings together predictions on single-letter changes across approximately 3 billion DNA letters and contains more than 1 petabyte of data.
At the core of the Atlas is the AlphaGenome model, which analyzes sequences of up to 1 million DNA letters and generates predictions about thousands of molecular features, such as gene expression, chromatin accessibility, and RNA processing. The model had outperformed general-purpose competitors in 25 of 26 variant-effect prediction tests. The Atlas aims to reduce the need to examine variants individually through an API and offers a resource more than 30 times larger than the AlphaFold Database. The data cover hundreds of human and mouse cell types and tissues, as well as more than 2 thousand 500 DNA motifs recurring throughout the genome.
DeepMind also introduced the AlphaGenome Variant Impact (AVI) score, which combines AlphaGenome and AlphaMissense predictions. The score, which can be used in coding and non-coding regions, contributed to the identification of a previously overlooked variant affecting the DNM1 gene in rare-disease research.
Why it matters
This resource could change prioritization in genomic analysis by enabling researchers to assess a broad set of possibilities within the same framework rather than examining variants one by one. Addressing not only coding regions but also non-coding regions associated with gene regulation makes it easier to study changes that fall outside approaches focused solely on protein-producing sequences. Predictions for different cell and tissue types provide researchers with a more detailed basis for comparison in cases where the biological effect of the same variant may vary depending on the context. However, how the predictions will be used in diagnosing rare diseases or validating research findings remains an open question that must be supported by experimental evidence. Their assumptions and margins of error should also be evaluated separately before clinical use.
Background
Google is not a new name in the FikirPilot archive: we have published 33 news articles mentioning the name in the past 90 days; the most recent was dated September 9, 2026.