Abstract:
Google DeepMind announced today (September 8, 2026) the launch of a new database "AlphaGenome Atlas", which uses artificial intelligence models to predict the impact of every possible single nucleotide variation in the human genome. It is called the most comprehensive catalog of how genetic mutations affect molecular biology to date.
The human genome consists of approximately 3 billion base pairs, but the functions of most of them remain a mystery. Scientists have a relatively deep understanding of the protein-coding regions that account for about 2% of the genome, but little is known about the remaining 98% of non-coding regions. Previously, DeepMind's AlphaGenome model has proven that a single change in these non-coding regions is enough to disrupt molecular processes such as protein synthesis, but the scientific community has always lacked a complete global picture.
To fill this gap, the research team used the AlphaGenome artificial intelligence model to calculate in advance the possible regulatory impacts of all 9 billion single-letter gene changes, thus generating a huge data set of up to 1PB (petabyte), which forms the core content of the AlphaGenome Atlas. In order to facilitate scientific researchers to quickly search and analyze from massive data, the team also specially designed the "AlphaGenome Variation Impact Score" (AVI Score). The score integrates the prediction results of coding and non-coding regions into a single, easy-to-use metric, allowing researchers to quickly identify the most valuable research directions without having to sift through thousands of pieces of data one by one.
Currently, AlphaGenome Atlas has demonstrated powerful auxiliary capabilities in multiple scientific research scenarios. In terms of research on rare genetic mutations, Laura Covill and her team at the Broad Institute used AVI scores to locate a key mutation in the DNM1 gene in a long-unsolved rare disease study. The tool predicted that this mutation would form an incorrect splicing site, providing key evidence support for finally solving this difficult case.
In the research of complex traits, it has always been difficult to identify rare non-coding variants related to complex traits due to the interference of statistical noise. Dr. Gareth Hawkes applied AlphaGenome Atlas to data from more than 54,000 participants in the UK Biobank and successfully identified 22% more non-coding gene associations by grouping variants according to predicted molecular effects. After further focusing on the top 1% of variants with the most significant impact, he identified 19 gene regions related to body mass index (BMI), which pointed out the direction for subsequent targeted research.
It is worth mentioning that AlphaGenome Atlas is currently open to the world through an intuitive and easy-to-use web portal that users do not need to have any programming skills to use, making this tool easily accessible to clinical researchers and biologists. Google DeepMind stated that this move is part of its ongoing commitment to accelerating genomics discovery and promoting inclusive science. The solid genomic insights provided by AlphaGenome Atlas are expected to further accelerate the pace of discovery in biological research.
Comments