SHELFCRITTER AI Digest

MODEL

AlphaGenome Atlas

modeltopic-notedeepmind

Overview

AlphaGenome Atlas is DeepMind‘s Sep 8 2026 release of a fully-precomputed catalog of predicted variant effects for all ~9 billion possible single-letter substitutions in the human genome — a distinctive shipping pattern from the earlier AlphaFold / AlphaGenome inference-on-demand shape. Rather than serving a model researchers query, DeepMind enumerated the entire query space up front and shipped the results, paired with a new AVI ranking score for prioritising candidate variants. The release pattern generalises to sequence-variant problems where the domain is combinatorially bounded, not to open-ended structure prediction.

Timeline

  • 2026-09-10-AI-Digest — DeepMind released AlphaGenome Atlas on Sep 8 as a precomputed catalog of predicted variant effects for all ~9 billion possible single-letter substitutions in the human genome, with a new AVI ranking score for prioritising candidate variants. Distinct release pattern from AlphaFold / AlphaGenome inference-on-demand — DeepMind enumerated the entire query space up front and shipped the results rather than serving a model researchers query.

Key Developments

  1. Predict-Everything-Up-Front Catalog Over Inference-On-Demand (September 8, 2026): The full catalog of predicted variant effects for all ~9 billion single-letter substitutions in the human genome is shipped precomputed rather than served on-demand — a different release pattern from the AlphaFold / AlphaGenome inference-on-demand shape. This pattern only works when the domain is combinatorially bounded — 9B single-letter substitutions is finite, protein conformation space is not — so the release shape generalises to sequence-variant problems, not to open-ended structure prediction.

  2. AVI Ranking Score as the Shipping Innovation: The AVI score is what turns a static catalog into a research tool — the ranking function that lets researchers prioritise candidate variants without re-querying a model. The corpus reads precomputed catalog + ranking score as the new DeepMind Life-Sciences release shape, likely to define the next wave of releases in domains where the query space is combinatorially bounded.

See also: DeepMind, Google, MOC - Major Companies.