google-deepmind/alphagenome_research

google-deepmind/alphagenome_research

Description: Research code accompanying AlphaGenome

Language: Python

License: Apache-2.0

Stars: 776

Forks: 129

Open issues: 1

Created: 2025-12-17T12:22:18Z

Pushed: 2026-06-03T13:33:23Z

Default branch: main

Fork: no

Archived: no

README: 

AlphaGenome Research

!Presubmit Checks

[Model Weights](#model-weights) | [Installation](#installation) | [Quick Start](#quick-start) | **Documentation** | **Community** | **Terms of Use**

AlphaGenome is a unified DNA sequence model designed to advance regulatory variant-effect prediction and shed light on genome function. It analyzes DNA sequences of up to 1 million base pairs to deliver predictions at single base-pair resolution across diverse modalities, including gene expression, splicing patterns, chromatin features, and contact maps.

This repository provides the following research code:

• An implementation of the AlphaGenome model, written in

JAX.

• An implementation of the

AlphaGenome API with accompanying variant scorers.

• A dataset loader for reading AlphaGenome training data from TFRecords.
• Colab notebooks for

running the model and analysing evaluations.

We strongly recommend using our AlphaGenome API to interact with the model without needing specialized hardware.

Installation

> [!TIP] > We strongly recommend you create a > Python Virtual Environment to > prevent conflicts with your system's Python environment.

To install, clone a local copy of this repository and run pip install:

$ git clone https://github.com/google-deepmind/alphagenome_research.git
$ pip install -e ./alphagenome_research

This will install any required dependencies, including this repository in development mode.

Model weights

To use our pre-trained model weights, you can download them from either:

• Kaggle or
• Hugging Face

Both require accepting our non-commercial model terms. Requests are processed immediately.

Model requirements

In order to run the model, we recommend running with at least an NVIDIA H100 GPU. Please ensure CUDA, cuDNN and JAX are correctly installed; the JAX installation documentation is a useful resource in this regard.

For training, we recommend running on Tensor Processing Units (TPUs) v3 or higher.

Quick start

The easiest way to interact with the AlphaGenome model is using the provided [DNA Model class](/src/alphagenome_research/model/dna_model.py). This wraps the core model and provides a more intuitive set of functions for creating predictions, scoring variants, performing in silico mutagenesis (ISM) and more.

It also provides the following factory functions to create a model instance using our pre-trained weights:

from alphagenome_research.model import dna_model

# To download from Kaggle:
model = dna_model.create_from_kaggle('all_folds')

# or Hugging Face:
model = dna_model.create_from_huggingface('all_folds')

Here's an example of making a variant prediction using model weights downloaded from Kaggle:

from alphagenome.data import genome
from alphagenome.visualization import plot_components
from alphagenome_research.model import dna_model
import matplotlib.pyplot as plt

model = dna_model.create_from_kaggle('all_folds')

interval = genome.Interval(chromosome='chr22', start=35677410, end=36725986)
variant = genome.Variant(
chromosome='chr22',
position=36201698,
reference_bases='A',
alternate_bases='C',
)

outputs = model.predict_variant(
interval=interval,
variant=variant,
ontology_terms=['UBERON:0001157'],
requested_outputs=[dna_model.OutputType.RNA_SEQ],
)

plot_components.plot(
[
plot_components.OverlaidTracks(
tdata={
'REF': outputs.reference.rna_seq,
'ALT': outputs.alternate.rna_seq,
},
colors={'REF': 'dimgrey', 'ALT': 'red'},
),
],
interval=outputs.reference.rna_seq.interval.resize(2**15),
# Annotate the location of the variant as a vertical line.
annotations=[plot_components.VariantAnnotation([variant], alpha=0.8)],
)
plt.show()

For further examples, please see our quick-start notebook.

Citing AlphaGenome

If you use AlphaGenome in your research, please cite using:

@article{alphagenome,
title={Advancing regulatory variant effect prediction with {AlphaGenome}},
author={Avsec, {\v Z}iga and Latysheva, Natasha and Cheng, Jun and Novati, Guido and Taylor, Kyle R. and Ward, Tom and Bycroft, Clare and Nicolaisen, Lauren and Arvaniti, Eirini and Pan, Joshua and Thomas, Raina and Dutordoir, Vincent and Perino, Matteo and De, Soham and Karollus, Alexander and Gayoso, Adam and Sargeant, Toby and Mottram, Anne and Wong, Lai Hong and Drot{\'a}r, Pavol and Kosiorek, Adam and Senior, Andrew and Tanburn, Richard and Applebaum, Taylor and Basu, Souradeep and Hassabis, Demis and Kohli, Pushmeet},
journal={Nature},
volume={649},
number={8099},
pages={1206--1218},
year={2026},
doi={10.1038/s41586-025-10014-0},
publisher={Nature Publishing Group UK London}
}

Acknowledgements

AlphaGenome's model release uses the following libraries and packages:

• Abseil
• anndata
• Chex
• Einshape

*…