AlphaGenome 2

AlphaGenome is a deep-learning model designed to predict how DNA sequence shapes gene activity and other cellular processes. It reads sequences up to one million DNA letters long and predicts thousands of genomic signals, including gene expression, chromatin accessibility, protein binding and RNA splicing. Many predictions are made at single-letter resolution, helping connect distant regulatory regions to the genes they may influence. The model was trained on human and mouse genomic data and brings many prediction tasks together in one system rather than relying on separate specialist models.

A key use is estimating how a genetic variant changes the predicted behavior of DNA: the model compares predictions for the original sequence and the sequence carrying the variant. In evaluations, it matched or exceeded specialized models on nearly all tested tasks, including important measures of splicing and gene regulation. This could help researchers interpret disease-associated variants, especially those outside protein-coding regions, and generate hypotheses about how they affect cells. AlphaGenome is a research tool for prioritizing and investigating candidate mechanisms, not a substitute for experimental validation or proof that a variant causes disease.