What it is
AlphaGenome is a single DNA sequence model that reads 1 megabase of surrounding context and predicts thousands of functional genomic measurements at single-base-pair resolution. In one pass it covers gene expression, chromatin accessibility, histone marks, transcription-factor binding, contact maps and splicing, dissolving the usual trade-off between how much sequence a model sees and how finely it resolves. Trained on human and mouse genomes, it matched or beat the strongest available specialist models on 25 of 26 variant-effect prediction evaluations.
Why it matters
Most disease-associated variation sits in the non-coding genome, and interpreting it has meant juggling many narrow models; AlphaGenome scores effects across thousands of genomic tracks at once. Scoring all modalities together let it reconstruct the mechanism of clinically relevant variants near the TAL1 oncogene, not merely flag them. Matching or exceeding specialist tools on 25 of 26 benchmarks in one unified model is a practical consolidation for variant interpretation.
Underlined numbers link to their source. Every metric and quoted figure is listed under Sources and data below.
Filed undergenomics, deep learning, gene regulation, variant effect prediction, DeepMind
Watch
A short explainer of this result.