Google DeepMind has launched AlphaGenome Atlas, a free ~1PB database that precomputes predictions for roughly 9 billion possible single-nucleotide variants (SNVs) across the human genome, removing the need for researchers to run the AlphaGenome model themselves via API. Built on the AlphaGenome model published in Nature (January 2026), the Atlas provides three layers of data: ~27,000 molecular-level scores per variant, an interpretable 18-feature AVI pathogenicity score with SHAP decomposition, and 2,601 clustered cis-regulatory motif groups. Benchmarks show strong wins in noncoding regions—for example, retrieval precision of 0.76 for intronic ClinVar pathogenic variants versus 0.44 for the best alternative, and 29.5% versus 12.5% recall for CADD on rare-disease candidates. Limitations include incomplete indel coverage, weaker 5' UTR performance, and a non-peer-reviewed companion preprint. Free for non-commercial use; commercial access pricing is not yet announced.
90 Billion 'What Ifs': DeepMind Precomputes Every Single-Letter Genome Edit in AlphaGenome Atlas
Summary
Google DeepMind has launched AlphaGenome Atlas, a free ~1PB database that precomputes predictions for roughly 9 billion possible single-nucleotide variants (SNVs) across the human genome, removing the need for researchers to run the AlphaGenome model themselves via API. Built on the AlphaGenome model published in Nature (January 2026), the Atlas provides three layers of data: ~27,000 molecular-level scores per variant, an interpretable 18-feature AVI pathogenicity score with SHAP decomposition, and 2,601 clustered cis-regulatory motif groups. Benchmarks show strong wins in noncoding regions—for example, retrieval precision of 0.76 for intronic ClinVar pathogenic variants versus 0.44 for the best alternative, and 29.5% versus 12.5% recall for CADD on rare-disease candidates. Limitations include incomplete indel coverage, weaker 5' UTR performance, and a non-peer-reviewed companion preprint. Free for non-commercial use; commercial access pricing is not yet announced.
This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178634663