Google DeepMind has expanded the capabilities of its genomic AI by launching AlphaGenome Atlas, a massive database that precomputes the molecular effects of every possible single nucleotide variant in the human genome. While previous efforts, such as those mapping genes linked to schizophrenia reported by AlexTech.ai, focused on specific networks or diseases, the Atlas provides a universal high-resolution map covering all 9 billion possible single-letter genetic changes.

Add AlexTech.ai asPreferred Source on Google

The dataset, totaling 1 petabyte, addresses a critical gap in genetics: the 98% of the human genome that does not code for proteins. While coding regions are well-understood, non-coding DNA often holds the key to how biological processes are regulated. By utilizing the AlphaGenome AI model, DeepMind has effectively tied cause and effect across the entire genome, allowing researchers to query predictions without needing to run complex models for every individual variant.

The AVI Score and Research Prioritization

To make this vast amount of data actionable, the Atlas introduces the AlphaGenome Variant Impact (AVI) score. This single metric combines predictions for both coding and non-coding regions, enabling scientists to rank variants by their likelihood of being deleterious or impactful. The AVI score is designed to filter out statistical noise, which is particularly prevalent when studying complex traits.

The utility of the AVI score has already been demonstrated in two key areas:

  • Rare Diseases: At the Broad Institute, researchers used the score to identify a critical variant in the DNM1 gene that created an incorrect splice site, solving a previously unsolved rare disease case.
  • Complex Traits: Using data from over 54,000 UK Biobank participants, Dr. Gareth Hawkes uncovered 22% more non-coding genetic associations and identified 19 genetic regions linked to body mass index (BMI).

Global Accessibility and Integration

The Atlas is available via an intuitive web portal requiring no coding skills, democratizing access for clinical biologists. For power users and bioinformaticians, the AVI scores have already been integrated into the Ensembl Variant Effect Predictor (VEP), and scripts for prioritizing variants are available through the science-skills repository.

This release marks a shift from targeted AI applications—like the general-purpose biomedical agents seen with Biomni—toward the creation of a foundational, precomputed reference for all human biology.