TL;DR: Google DeepMind launched AlphaGenome Atlas on September 8, 2026, a 1-petabyte catalog with predicted molecular effects for all 9 billion possible single-letter changes in human DNA, plus a single-number ranking score to help researchers prioritize variants.

What launched on September 8

Google DeepMind introduced AlphaGenome Atlas on September 8, 2026, as a platform containing predictions for the effects of 9 billion single-nucleotide variants, or every single-letter change possible in the human genome, according to Google DeepMind.

The organization described the release as the most comprehensive catalogue of how genetic mutations affect molecular biology, according to Google DeepMind.

The Verge reported the launch on Sept. 8, 2026, at 2:00 PM UTC, describing AlphaGenome Atlas as a predictive map of every possible DNA letter change in the human genome, according to The Verge.

The human genome contains roughly three billion letter pairs written in four chemical letters, usually shortened to A, C, G, and T, according to The Verge.

Those letters carry instructions for how, when, and where genes are switched on and off, and to what degree, according to The Verge.

Changes to individual letters can be harmless, contribute to ordinary differences between people, or play a role in disease, and identifying which changes matter is a major challenge because there are roughly nine billion potential single-letter substitutions, according to The Verge.

Google DeepMind said testing each of the roughly 9 billion possible single-letter mutations in the laboratory is practically impossible, according to Google DeepMind.

Why the focus is on non-coding DNA

Scientists understand the 2% of the human genome that codes for proteins relatively well, but have only limited knowledge of the remaining 98%, according to Google's announcement on the Google Blog.

The 2% figure refers to coding regions that encode proteins, while the remaining 98% orchestrates gene activity and houses most trait-associated variants, according to Google DeepMind.

AlphaGenome Atlas is designed to work for both coding regions and non-coding regions, according to Google DeepMind.

That distinction matters because much of the genome does not directly code for proteins but can control how genes behave, according to The Verge.

The portion of the human genome that encodes proteins is less than 3 percent in another description of the same landscape, with most of the genome described as non-coding, according to Ars Technica.

That non-coding portion includes centromeres that help ensure chromosomes are divided evenly between cells, caps that protect chromosome ends, regulatory DNA that controls gene activity, signals that help determine what should and should not be included in mature messenger RNAs, and sequences that help control DNA packaging, according to Ars Technica.

Most non-coding DNA is described as remnants of ancient viral infections, DNA-level parasites, genes inactivated by mutation and related material, and determining what is useful has remained an ongoing challenge, according to Ars Technica.

DNA-binding proteins are not highly selective about sequences, can potentially bind at random locations, tolerate mutation, are often cell-type specific, and frequently act through many compact binding sites rather than one site alone, according to Ars Technica.

How AlphaGenome became Atlas

The project builds on AlphaGenome, an artificial intelligence model DeepMind unveiled last year to help scientists identify genetic drivers of disease, as well as AlphaMissense, an earlier tool focused on predicting which small mutations might alter proteins, according to The Verge.

AlphaGenome is an AI model that can predict how genetic variants impact biological processes, and it has found widespread use for analyzing specific variants, according to Google DeepMind.

AlphaGenome was trained using public databases of human and mouse genomes, allowing it to learn patterns between DNA changes and biological processes, according to The Verge.

The system is currently limited to sequences from mice and humans and was trained on a limited number of exhaustively studied cell types, according to Ars Technica.

For researchers studying a gene with changes in non-coding regions, AlphaGenome can provide a hint about significance along with a hypothesis about why, and its predictions are generally as good as or better than those from specialized software tools, according to Ars Technica.

In a press briefing, DeepMind genomics lead Žiga Avsec acknowledged that AlphaGenome had already been released but said creating the genome-wide catalog took time, according to The Verge.

“Basically it took us some time to really precompute and also analyze this many variants because the space is so big,” Avsec said in that briefing, according to The Verge.

By precomputing AlphaGenome's predictions at scale, DeepMind created a resource intended to expand the model's reach and give researchers a big-picture view across the genome, according to Google DeepMind.

Applying those predictions to billions of possible variants produced a dataset Google says is roughly 1 petabyte in size, according to The Verge.

A 1-petabyte catalog with thousands of signals per variant

Google DeepMind said AlphaGenome Atlas is a massive 1-petabyte dataset, more than 30 times larger than the AlphaFold Database, according to Google DeepMind.

When DeepMind expanded the AlphaFold Database in 2022, it grew available 3D structure information from around 190,000 experimental structures to more than 200 million structure predictions, covering nearly all catalogued proteins known to science, according to Google DeepMind.

That database provided a portal for researchers with no coding experience, with visualizations and support for large-scale protein structure analysis, and became a resource for life-sciences discovery, according to Google DeepMind.

AlphaGenome Atlas follows the same accessibility goal by giving scientists an intuitive way to explore a vast dataset, according to Google DeepMind.

For each variant, Atlas contains thousands of molecular effect predictions across multiple aspects of gene regulation, spanning hundreds of human and mouse cell types and tissues, according to Google DeepMind.

Those predictions cover gene expression, transcription initiation, chromatin accessibility, histone modifications, transcription factor binding, chromatin contact maps, splice site usage, and splice junction coordinates and strength, according to Ars Technica.

In practical terms, one prediction might concern expected transcription changes in a cell context, altered splice-site usage in neurons, or weakened transcription-factor binding in a particular cellular context, reflecting a predicted molecular mechanism rather than only a generic score, according to the Switzerland-based explainer cited alongside the launch materials.

The AVI score for ranking 9 billion possibilities

To help researchers sift through billions of possibilities, Google is also releasing what it calls an AlphaGenome Variant Impact score, or AVI, that draws on other models for predicting DNA-change effects, according to The Verge.

“Now, researchers can rapidly rank variants and interpret their molecular effects at the same time,” the company said in its blog, according to The Verge.

The AVI combines AlphaGenome with AlphaMissense, DeepMind's model for predicting the impact of protein-altering DNA variants, condensing both models' predictions into a single number, according to Google DeepMind.

The AVI score is a single number describing the impact for each genetic variant, according to Google DeepMind.

Each AVI score is linked to distinct biological features driving it, such as aspects of gene regulation predicted by AlphaGenome or the protein-impact score from AlphaMissense, according to Google DeepMind.

Google DeepMind said its testing shows the AVI score provides best-in-class performance across many variant pathogenicity and rare-disease benchmarks, according to Google DeepMind.

To help interpret scores, DeepMind also calculated AVI feature attributions that highlight which molecular processes, such as RNA splicing or gene expression, are predicted to be most disrupted by each variant, according to Google DeepMind.

The Atlas also includes a comprehensive collection of over 2,500 recurrent DNA sequences, described as the words of the genome, and their locations, according to Google DeepMind.

Together, the molecular-effect predictions, AVI score, feature attributions and motif collection are intended to support tasks from rapid variant ranking to detailed investigation of variant function, according to Google DeepMind.

Early test: an unsolved rare disease

At the Broad Institute, Laura Covill and her team used the AVI score to prioritize variants for unsolved rare-disease research, according to the Google Blog.

The tool highlighted a critical variant in the DNM1 gene, predicting that it created an incorrect splice site, and that provided supporting evidence to successfully solve the case, according to the Google Blog.

External collaborators already used AlphaGenome Atlas to identify and experimentally verify key variants in unsolved rare-disease research and to find rare variants associated with common traits, according to Google DeepMind.

Early test: 54,000 genomes and body-mass traits

Dr. Gareth Hawkes applied AlphaGenome Atlas to data from more than 54,000 UK Biobank participants, according to the Google Blog.

Identifying rare, non-coding variants linked to complex traits is difficult because of statistical noise, according to the Google Blog.

By grouping variants based on predicted molecular effects, Hawkes uncovered 22% more non-coding genetic associations, according to the Google Blog.

Focusing on the top 1% of impactful variants, he identified 19 genetic regions linked to body mass index, directing the next stage of targeted research, according to the Google Blog.

The University of Exeter reported the same program on September 10, 2026, saying Hawkes applied Atlas to whole-genome data from over 54,000 UK Biobank participants and made elusive genetic signals more identifiable, according to the University of Exeter.

The university said grouping rare variants by predicted molecular effects uncovered 22 per cent more non-coding associations that otherwise would not have been detectable in statistical noise, according to the University of Exeter.

That work also pinpointed specific regulatory variants driving abundance of critical circulating proteins, including PLA2G7 linked to aging and EGLN1 described as a vital cellular oxygen sensor, according to the University of Exeter.

Looking at hundreds of millions of non-coding variants in the UK Biobank for possible links to body mass index, and focusing on the 1 per cent predicted to be most impactful, Hawkes identified genetic regions that could help direct targeted research into that trait, according to the University of Exeter.

“The smaller we can make the haystack, the more likely we are to find the needle,” Hawkes said, according to the University of Exeter.

How researchers can access it

AlphaGenome Atlas is available through an intuitive website portal that requires zero coding skills, a design intended to expand access for clinical researchers and biologists, according to the Google Blog.

AlphaGenome Atlas is available through a website portal, an AlphaGenome programming interface, and as a skill in Google Antigravity, according to Google DeepMind.

Google says scientists can explore predictions through a web portal, as a skill in its agentic development platform Antigravity, and through its AlphaGenome interface, according to The Verge.

The company is making Atlas available to researchers for noncommercial use through its website starting September 8, with commercial use on Google Cloud planned as “soon,” according to The Verge.

Place in Google's science push

Atlas is the latest in a series of Google efforts to use AI for science and medicine, arriving as DeepMind cofounder Demis Hassabis steps back from running the AI lab to focus on scientific research, including leading drug-discovery spinoff Isomorphic Labs, according to The Verge.

The company's best-known work in the area is AlphaFold, the protein-structure prediction model that won Hassabis and John Jumper the 2024 Nobel Prize in Chemistry, according to The Verge.

Google has also developed AI tools for weather prediction, computing and mathematics solutions, and an agentic co-scientist for researchers, according to The Verge.

Sources referenced