Awesome

EMDS

This repository provides the implemention for the paper Equivariant Score-based Generative Diffusion Framework for 3D Molecules.

Please cite our paper if our datasets or code are helpful to you ~ 😊

Requirements

• Python 3.7
• pytorch 1.12.0
• rdkit 2022.3.5
• pyparsing 2.4.7

Dataset

• QM9: https://deepchemdata.s3-us-west-1.amazonaws.com/datasets/gdb9.tar.gz

Running Experiments

Preparations

To preprocess the 3D molecular dataset QM9 for training model, run the following command:

python data/preprocess_3d.py
python data/split_generators.py

Configs

The configurations are provided on the config/ directory in YAML format.

Training

CUDA_VISIBLE_DEVICES=0 python main.py --type train --config train --seed 42

Generation and Evaluation

CUDA_VISIBLE_DEVICES=0 python main.py --type sample --config sample

Acknowledgements

EMDS builds upon the source code from the projects GDSS, G-SphereNet and EDM.

We thank their contributors and maintainers!

Citation

Please cite our paper if our datasets or code are helpful to you.

H. Zhang, Y. Liu, X. Liu, C. Wang, and M. Guo, "Equivariant Score-based Generative Diffusion Framework for 3D Molecules", BMC Bioinformatics 25, 203 (2024), DOI: 10.1186/s12859-024-05810-w