AlphaGenome

AlphaGenome is a deep-learning model developed by Google DeepMind for predicting how DNA sequences regulate gene expression and how single-nucleotide variants may disrupt that regulation. It was launched on 25 June 2025.

The model takes DNA sequences of up to one million base pairs as input and produces thousands of quantitative predictions characterising regulatory activity, including gene-expression levels, chromatin accessibility, three-dimensional genome contacts and RNA splice junctions. It can also score the effects of variants by comparing predictions for mutated and unmutated sequences.

Its architecture combines convolutional layers that detect short patterns in the sequence, transformers that share information across all positions, and prediction layers that output 11 modalities concurrently. The model was trained on human and mouse genomic data (ENCODE, GTEx, 4D Nucleome, FANTOM5). AlphaGenome is complementary to AlphaMissense and builds on DeepMind's earlier Enformer model. It is available in preview through the AlphaGenome API for non-commercial research use.