AlphaGenome Atlas
An interactive genomic variant atlas tool developed by Google DeepMind, providing systematic coverage of predicted effects for approximately nine billion DNA variants. Accessible freely via a web browser without requiring programming skills or a local computing environment, and is available to academic researchers.
Core Facts
Timeline (last 90 days)
AlphaGenome Atlas provides predictions rather than conclusions; its accuracy is limited by the coverage of training data and the boundaries of model capability, and clinical decision-making still requires experimental validation.
Multiple research institutions including the Broad Institute and the University of Exeter have begun using the AlphaGenome Atlas tool to identify potentially pathogenic DNA variants and interpret their roles in disease.
Information currently publicly available about AlphaGenome Atlas comes mainly from the tool developer's introductory release, and specific performance metrics, the scale of validation datasets, and comparative results against existing methods have not been fully disclosed.
Functional predictions of variants made by AI still require experimental validation, and the ability of models to generalize to rare variants outside the training data distribution and to genomes from diverse populations remains questionable.
The value of AlphaGenome Atlas lies not only in identifying potentially problematic variant sites, but also in explaining the functional consequences brought about by those variants.
By grouping variants according to their predicted molecular effects, Gareth Hawkes found a 22% increase in non-coding genetic associations.
Hawkes used Atlas to investigate associations with BMI, focusing on the top 1% of non-coding variants by predicted impact, and identified 19 genetic regions for further study.
The figure of 9 billion comes from the genome's approximately 3 billion positions, each of which can be substituted by one of the other three letters, giving 3 billion × 3 ≈ 9 billion possible single-base changes.
AlphaGenome Atlas is both a data resource and a website, and has built a genome browser for users to view genomic regions from different perspectives.
AlphaGenome Atlas is described by the team as the most comprehensive and accurate variant effect prediction platform currently available, with a data scale of approximately 1 PB.
24 more timeline events
All Facts (20)
The AlphaGenome Atlas contains approximately 1 PB (1,000 TB) of data.
65%VerifiedAlphaFold solves protein structure prediction, while AlphaGenome Atlas further answers what functional consequences genetic changes will lead to.
65%UnverifiedAlphaGenome Atlas is an interactive resource that systematically maps the predicted effects of all approximately 9 billion DNA variants.
70%UnverifiedAlphaGenome Atlas pre-computed AlphaGenome model predictions for all 9 billion possible mutations, resulting in a dataset of 1 PB (1,000 TB) in scale.
60%UnverifiedAlphaGenome Atlas can systematically analyze and predict the effects that genetic mutations may have on the human body, without the need for time-consuming experimental validation.
60%UnverifiedAlphaGenome Atlas is a platform covering a prediction map of every possible DNA letter change in the human genome.
60%UnverifiedAlphaGenome Atlas provides predictions rather than conclusions; its accuracy is limited by the coverage of training data and the boundaries of model capability, and clinical decision-making still requires experimental validation.
50%UnverifiedFunctional predictions of variants made by AI still require experimental validation, and the ability of models to generalize to rare variants outside the training data distribution and to genomes from diverse populations remains questionable.
50%UnverifiedThe value of AlphaGenome Atlas lies not only in identifying potentially problematic variant sites, but also in explaining the functional consequences brought about by those variants.
50%UnverifiedHawkes used Atlas to investigate associations with BMI, focusing on the top 1% of non-coding variants by predicted impact, and identified 19 genetic regions for further study.
50%UnverifiedThe figure of 9 billion comes from the genome's approximately 3 billion positions, each of which can be substituted by one of the other three letters, giving 3 billion × 3 ≈ 9 billion possible single-base changes.
50%UnverifiedBy grouping variants according to their predicted molecular effects, Gareth Hawkes found a 22% increase in non-coding genetic associations.
50%UnverifiedAlphaGenome Atlas is described by the team as the most comprehensive and accurate variant effect prediction platform currently available, with a data scale of approximately 1 PB.
50%UnverifiedAlphaGenome Atlas is both a data resource and a website, and has built a genome browser for users to view genomic regions from different perspectives.
50%UnverifiedWhen using the Atlas, focusing on specific molecular phenotype categories and their corresponding confidence intervals is more helpful for assessing reference value than looking directly at composite scores.
50%UnverifiedAlphaGenome Atlas is freely available to academic researchers.
50%UnverifiedAlphaGenome Atlas covers the theoretically complete variant space, rather than only known variants observed in populations.
50%UnverifiedAlphaGenome Atlas provides predicted effects rather than experimentally validated definitive conclusions, and its predictions should serve as a starting point for research hypotheses.
50%UnverifiedAlphaGenome Atlas can be accessed directly in a standard web browser without writing code or setting up a local computing environment.
50%UnverifiedAlphaGenome Atlas has constructed a systematic resource map covering multiple tissue types, multiple cell types, and multiple molecular readouts.
50%