Bayer-Group / eqgat

Research repository for the proposed equivariant graph attention network that operates on large biomolecules proposed by Le et al. (2022)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Representation Learning on Biomolecular Structures using Equivariant Graph Attention

Pytorch implementation for the manuscript Representation Learning on Biomolecular Structures using Equivariant Graph Attention presented at the Machine Learning For Structural Biology Workshop at NeurIPS 2022 (short paper) as well as in the Learning On Graphs Conference 2022 as full-length conference paper.

Overview

Here we provide benchmark scripts for our experiments on the EQGAT architecture. Make sure to install the eqgat library.

git clone https://github.com/Bayer-Group/eqgat.git
cd eqgat

This repository is organised as follows:

  • eqgat/ contains the implementation of the Equivariant Graph Attention Model with all required submodules. Additionally, we provide implementations of other recent 3D Graph Neural Networks.
  • experiments/ contains the 5 python training-scripts from the ATOM3D and 1 synthetic datasets. To execute each training script, please refer to the corresponding README.md in the sub-directories.

Installation with GPU support

# install the conda environment
conda env create -f environment.yml 
conda activate eqgat
pip install -e .

Experiments

All experiments presented in the paper can be found in the experiments/ directory.
Make sure to download all requested public datasets from ATOM3D as described in the corresponding READMEs.

Example

A minimal example using the proposed SO(3) equivariant graph attention network can be found in eqgat/README.md

License

Code is available under BSD 3-Clause License.

Reference

If you make use of our model architecture, please cite our full-length manuscript:

T. Le et al., Representation Learning on Biomolecular Structures using Equivariant Graph Attention. Proceedings of the First Learning on Graphs Conference (LoG 2022), PMLR 198, Virtual Event, December 9–12, 2022.

@inproceedings{
le2022representation,
title={Representation Learning on Biomolecular Structures using Equivariant Graph Attention},
author={Tuan Le and Frank Noe and Djork-Arn{\'e} Clevert},
booktitle={Learning on Graphs Conference},
year={2022},
url={https://openreview.net/forum?id=kv4xUo5Pu6}
}

About

Research repository for the proposed equivariant graph attention network that operates on large biomolecules proposed by Le et al. (2022)

License:BSD 3-Clause "New" or "Revised" License


Languages

Language:Python 50.9%Language:Jupyter Notebook 49.1%