Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/minkaixu/egno
https://github.com/minkaixu/egno
dynamic-simulations geometric-deep-learning graph-neural-networks molecular-dynamics neural-operator
Last synced: 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/minkaixu/egno
- Owner: MinkaiXu
- License: mit
- Created: 2024-02-05T08:13:16.000Z (12 months ago)
- Default Branch: main
- Last Pushed: 2024-03-27T20:02:10.000Z (10 months ago)
- Last Synced: 2024-03-28T18:34:39.327Z (10 months ago)
- Topics: dynamic-simulations, geometric-deep-learning, graph-neural-networks, molecular-dynamics, neural-operator
- Language: Python
- Homepage:
- Size: 10.3 MB
- Stars: 9
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Equivariant Graph Neural Operator for Modeling 3D Dynamics
[](https://github.com/MinkaiXu/EGNO/blob/main/LICENSE)
[](https://arxiv.org/abs/2401.11037)
This repository contains the official implementation of our [Equivariant Graph Neural Operator](https://arxiv.org/abs/2401.11037).
> [**Equivariant Graph Neural Operator for Modeling 3D Dynamics**](https://arxiv.org/abs/2401.11037)
> Minkai Xu*, Jiaqi Han*, Aaron Lou, Jean Kossaifi, Arvind Ramanathan, Kamyar Azizzadenesheli, Jure Leskovec, Stefano Ermon, Anima Anandkumar
>
**Stanford University, NVIDIA, Argonne National Laboratory, California Institute of Technology**
## Environment
You can install the exact environment with `env.yml`:
```bash
conda env create -f env.yml
```
or manually install the following packages:
```
python=3.8.17
pytorch=2.0.1
scipy=1.10.1
```
You may also need `mdanalysis` if you want to process the protein MD data.
## Data Preparation
**1. Simulation dataset**
We provide the data preprocessing code in `simulation/dataset`. One can simply run
```python
cd simulation/dataset
python -u generate_dataset.py
```
**2. Motion capture dataset**
We provide our pre-processed dataset as well as the splits in `motion/dataset` folder.
**3. MD17 dataset**
We provide the splits in `md17` folder. The dataset can be downloaded from [here](http://www.sgdml.org/#datasets) and then placed in `md17` folder.
**4. Protein MD**
We provide the data preprocessing code in `mdanalysis/preprocess.py`. One can simply run
```python
python mdanalysis/preprocess.py
```
after setting the correct data path specified as the variable `tmp_path` in `preprocess.py`.
## Train the EGNO
**1. Simulation dataset**
```bash
python -u main_simulation_simple_no.py --config_by_file --outf $log_dir
```
**2. Motion capture**
```bash
python -u main_mocap_no.py --config_by_file --outf $log_dir
```
**3. MD17**
```bash
python -u main_md17_no.py --config_by_file --outf $log_dir
```
**4. Protein MD**
```bash
python -u main_mdanalysis_no.py --config_by_file --outf $log_dir
```
## Evaluation
All evaluations (validation and testing) are conducted along with training.
## Citation
Please consider citing the our paper if you find it helpful. Thank you!
```
@article{xu2024equivariant,
title={Equivariant Graph Neural Operator for Modeling 3D Dynamics},
author={Xu, Minkai and Han, Jiaqi and Lou, Aaron and Kossaifi, Jean and Ramanathan, Arvind and Azizzadenesheli, Kamyar and Leskovec, Jure and Ermon, Stefano and Anandkumar, Anima},
journal={arXiv preprint arXiv:2401.11037},
year={2024}
}
```
## Contact
If you have any question, welcome to contact me at:
Minkai Xu: [email protected]
## Acknowledgement
This repo is built upon several great codebases, including [EGNN](https://github.com/vgsatorras/egnn) and [GMN](https://github.com/hanjq17/GMN). We thank the authors for their great work and open-sourcing the code!