README.md

CHEM

Official implementation of CHEM: Causally and Hierarchically Explaining Molecules using PyTorch.

NOTE: This repository supports WandB MLOps platform!

---

Arguments

Argument	Description	Options
--dataset	Dataset to use	mutag, bbbp, bace, clintox, tox21, sider, syn
--target_col	For multi-task dataset, assign a single task.	clintox: 0, 1, tox21: 0 ~ 12, sider: 0 ~ 26
--bias	For synthetic dataset syn, assign bias of data.	0.5, 0.7, 0.9

---

Imputation & Evaluation

Run the following command to perform train and evaluation:

python main.py --dataset mutag
python main.py --dataset bbbp
python main.py --dataset bace
python main.py --dataset clintox --target_col 0
python main.py --dataset tox21 --target_col 0
python main.py --dataset sider --target_col 0
python main.py --dataset syn --bias 0.5

Citation

If you use this code in your research, please cite:

@inproceedings{woo2025chem,
  title={CHEM: Causally and Hierarchically Explaining Molecules},
  author={Woo, Gyeongdong and Cho, Soyoung and Kim, Donghyeon and Na, Kimoon and Kim, Changhyun and Choi, Jinhee and Jeon, Jong-June},
  booktitle={Proceedings of the 34th ACM International Conference on Information and Knowledge Management},
  pages={3323--3332},
  year={2025}
}