google-deepmind/alphagenome

What it solves

AlphaGenome is designed to decipher the regulatory code within DNA sequences. It solves the problem of predicting how changes in DNA (variants) affect biological functions, providing a high-resolution map of the genome's regulatory landscape.

How it works

It is a multimodal model that analyzes DNA sequences up to 1 million base pairs long. It generates predictions for various functional outputs, including gene expression, splicing patterns, chromatin features, and contact maps, often at single base-pair resolution. The project also provides the "Atlas" dataset, which contains pre-computed variant effect scores (AlphaGenome Variant Impact or AVI scores) and feature importances across the entire human genome.

Who it’s for

Researchers in genomics and bioinformatics who need to perform theoretical modeling and research on genomic variants and their regulatory effects.

Highlights

• Multimodal Predictions: Predicts gene expression, splicing, and chromatin features from DNA sequences.
• High Resolution: Delivers predictions at single base-pair resolution for most outputs.
• Large-scale Analysis: Provides the Atlas dataset for pre-computed variant effect scores across the human genome.
• State-of-the-art Performance: Achieves high performance across genomic prediction benchmarks and variant effect prediction tasks.