A searchable database maps predicted molecular consequences for all possible single-letter DNA swaps across the human genome, accessible through a web portal for non-commercial use and built from DeepMind’s AlphaGenome model.
AlphaGenome Atlas provides predictions for the three possible single-letter substitutions at every DNA position, producing a petabyte-scale searchable map.
The atlas brings tissue-specific, molecular-level predictions (gene activity, chromatin changes) to a browser-based portal, lowering the coding barrier that existed with the AlphaGenome API.
AlphaGenome creates predictions by detecting short DNA patterns, sharing information across larger regions, and translating patterns into downstream biological effects.
# Summary DeepMind released the AlphaGenome Atlas, a comprehensive, browser-accessible database that predicts the molecular effects of every possible single-letter DNA substitution across the human genome. The atlas builds on the AlphaGenome AI, translating its genome-scale predictions into a searchable web portal intended for non-commercial research use.
# What the atlas contains The atlas stores predictions for all three possible substitutions at each genomic base (for example, A→T, A→C, A→G). Generating that output produced roughly a petabyte of data. Each substitution comes with thousands of model-derived predictions about molecular consequences in different tissues, including changes in nearby gene activity and effects on chromatin structure.
# Why this matters now Until AlphaGenome, analysis required access via an API and coding skills. Many researchers used the model, but the requirement to program limited access. The AlphaGenome Atlas aims to open those predictions to researchers who prefer a browser interface, enabling easier lookup of variants without writing code.
# How AlphaGenome makes its predictions AlphaGenome works in three stages. First it identifies short sequence patterns in DNA. Then it propagates that local information across larger genomic regions so the model can connect local motifs to distant letters. Finally it translates the combined information into predictions about downstream molecular effects. The model is customizable for different research questions.
# What researchers can do with the atlas
# Access and attribution DeepMind published a paper describing the work and posted details in a blog post. Pushmeet Kohli, DeepMind's vice president of science, said the atlas allows any researcher to access a comprehensive map of the genome and its variations through a browser. The portal is for non-commercial use.
# Open questions and next steps The original report notes that interpreting the atlas remains a task for researchers. The atlas supplies model predictions at scale, but using those predictions to establish causation, validate mechanisms, or design therapeutic edits requires experimental follow-up and domain-specific analysis.
# Bottom line AlphaGenome Atlas centralizes model predictions for every single-base DNA substitution, surfaces tissue-specific molecular effects, and removes a coding barrier for many researchers. It represents a resource for prioritizing variants across coding and non-coding genome regions, while leaving interpretation and experimental validation to the research community.
Open the app view to save this story, compare related coverage, and continue from the same source.