Equivariant Graph Neural Operator for Modeling 3D Dynamics

This repository contains the official implementation of our Equivariant Graph Neural Operator.

Equivariant Graph Neural Operator for Modeling 3D Dynamics
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:

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

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 and then placed in md17 folder.

4. Protein MD

We provide the data preprocessing code in mdanalysis/preprocess.py. One can simply run

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

python -u main_simulation_simple_no.py --config_by_file --outf $log_dir

2. Motion capture

python -u main_mocap_no.py --config_by_file --outf $log_dir

3. MD17

python -u main_md17_no.py --config_by_file --outf $log_dir

4. Protein MD

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: minkai@cs.stanford.edu

Acknowledgement

This repo is built upon several great codebases, including EGNN and GMN. We thank the authors for their great work and open-sourcing the code!

MinkaiXu/EGNO