https://github.com/bayer-group/eqgat
Research repository for the proposed equivariant graph attention network that operates on large biomolecules proposed by Le et al. (2022)
https://github.com/bayer-group/eqgat
deeplearning graphneuralnetwork neural-network protein-structure
Last synced: 4 months ago
JSON representation
Research repository for the proposed equivariant graph attention network that operates on large biomolecules proposed by Le et al. (2022)
- Host: GitHub
- URL: https://github.com/bayer-group/eqgat
- Owner: Bayer-Group
- License: bsd-3-clause
- Created: 2022-11-28T19:38:39.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2022-12-01T10:38:45.000Z (almost 3 years ago)
- Last Synced: 2025-04-12T13:49:12.075Z (6 months ago)
- Topics: deeplearning, graphneuralnetwork, neural-network, protein-structure
- Language: Python
- Homepage:
- Size: 455 KB
- Stars: 19
- Watchers: 4
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Codeowners: CODEOWNERS
Awesome Lists containing this project
README
## 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**](https://www.mlsb.io/) (short paper)
as well as in the [**Learning On Graphs Conference 2022**](https://logconference.org/) 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](https://www.atom3d.ai/) 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](https://www.atom3d.ai/) 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}
}
```