Google DeepMind has launched AlphaGenome Atlas, a new resource designed to help researchers explore how genetic variation may affect molecular biology across the human genome.

The Atlas contains precomputed predictions for the effects of 9 billion single-nucleotide variants, covering every possible single-letter change in the human genome.

Built using predictions from AlphaGenome, the resource provides information on how individual variants may influence biological processes including gene expression, RNA splicing and chromatin accessibility, across hundreds of human and mouse cell types and tissues.

Importantly, the Atlas covers both coding and non-coding regions of the genome, making it potentially useful for investigating variants whose functional consequences can be particularly difficult to interpret.

From billions of variants to interpretable predictions

One of the new features introduced alongside the Atlas is the AlphaGenome Variant Impact (AVI) score.

The score combines predictions from AlphaGenome with AlphaMissense, Google DeepMind's model for predicting the effects of protein-altering variants, into a single measure intended to help researchers rapidly rank variants according to their predicted impact.

Researchers can then go beyond the score to explore which biological features contribute to a prediction, including effects related to splicing, gene expression, chromatin accessibility and protein function.

The Atlas also includes a collection of more than 2,500 DNA sequence motifs, providing another layer for investigating the regulatory mechanisms potentially affected by genetic variation.

From prediction to biological research

Google DeepMind reports that external research collaborators have already started applying the resource to questions in rare disease genetics, population genetics and gene regulation.

In one example, researchers investigating unsolved rare diseases used the AVI score to prioritise candidate variants and identified a variant affecting DNM1, a gene associated with epileptic encephalopathy. AlphaGenome predicted that the variant created an incorrect splice site, a mechanism that was subsequently supported by experimental screening.

In another application, researchers analysed whole-genome data from more than 54,000 UK Biobank participants. By grouping rare variants according to their predicted molecular effects, they identified additional associations involving non-coding variants that had previously been difficult to detect.

These examples also highlight an important point: computational predictions can help researchers prioritise hypotheses and experiments, but experimental validation remains essential for establishing biological effects.

Explore AlphaGenome Atlas

AlphaGenome Atlas is now freely available for non-commercial research through an online portal, allowing researchers to explore its predictions without needing to work directly with the underlying model or dataset.

Researchers can also access AlphaGenome programmatically through its API.

🔗 Explore AlphaGenome Atlas: https://alphagenome.google/atlas

If you work in genomics, molecular biology, bioinformatics or a related field, we encourage you to explore the Atlas and see how it could support your research.

Have you already tried AlphaGenome Atlas? Share your experience in the comments. We would be interested to hear how researchers in the FEBS community are using it, what you find useful and where you see its limitations.