google-deepmind/alphagenome_research

Research code accompanying AlphaGenome

What it solves

AlphaGenome addresses the challenge of predicting the effects of regulatory genetic variants on genome function. It provides a way to understand how changes in DNA sequences impact various biological processes at a single base-pair resolution.

How it works

It is a unified DNA sequence model implemented in JAX that analyzes sequences up to 1 million base pairs long. The model predicts multiple modalities of genomic data, including gene expression, splicing patterns, chromatin features, and contact maps. Users can interact with the model via a Python API to create predictions, score variants, and perform in silico mutagenesis (ISM).

Who it’s for

This project is designed for genomics researchers and scientists studying regulatory variant-effect prediction and genome function.

Highlights

  • High Resolution: Delivers predictions at single base-pair resolution.
  • Large Context Window: Capable of analyzing DNA sequences up to 1 million base pairs.
  • Multi-modal Predictions: Predicts gene expression, splicing, chromatin features, and contact maps.
  • Flexible Deployment: Offers both a research code implementation for local use (requiring high-end GPUs/TPUs) and a managed API for easier access.

Related

  • Project
    google-deepmind/alphagenome

    A unifying model for deciphering the regulatory code within DNA sequences, providing multimodal predictions for gene expression, splicing, and chromatin features.

  • Project
    songlab-cal/gpn

    A suite of genomic language models for predicting the effects of genome-wide variants using unaligned or aligned genomic sequences.

  • Project
    pinellolab/DNA-Diffusion

    A diffusion-based generative model for creating cell-type specific synthetic regulatory DNA sequences of 200bp.

  • Project
    aristoteleo/dynamo-release

    Dynamo is an analytical framework for mapping transcriptomic vector fields in single-cell RNA-seq data to predict cell fate transitions and regulatory mechanisms.

  • Project
    OmicsML/dance

    A deep learning toolkit for single-cell gene expression analysis that provides a benchmark platform for reproducing experiments and automated preprocessing recommendations.