🚀 Introduction

Our Polyconf achieves state-of-the-art performance in polyconf conformation generation. In particular, our PolyConf decompose the polymer conformation into a series of local conformations (i.e., the conformations of its repeating units), generating these local conformations through an autoregressive model, and then generating their orientation transformations via a diffusion model to assemble them into the complete polymer conformation, thereby effectively accommodating their unique structural characteristics.

⚒️ Environment

The required packages have been listed in requirements.txt. To set up your environment, please execute:

pip install -r requirements.txt

📦 Dataset

The processed dataset has been provided in this link, please download this dataset and organize the ./dataset directory as follows:

dataset
├── true_confs
├── dict.txt
├── test.lmdb
├── valid.lmdb
├── train.lmdb
├── test_data_index.csv

💪 Experiments

Training

Our model weight has been provided in this link. If using ours, please place it in the ./phase2_ckpt folder and rename it to checkpoint_best.pt.

Of course, you can also train from scratch by running the following scripts.

bash train_phase1.sh
bash train_phase2.sh

Inference

bash inference.sh

Evaluation

python extract_confs.py
python eval.py 

👍 Acknowledgments

This code is built upon Uni-Mol, Uni-Core, MAR, MolCLR, TorsionalDiff and FrameDiff. Thanks for their contribution.

📌 Citation

If you find this work useful for your research, please consider citing it. 😊

@inproceedings{wang2025polyconf,
  title={PolyConf: Unlocking Polymer Conformation Generation through Hierarchical Generative Models},
  author={Wang, Fanmeng and Guo, Wentao and Ou, Qi and Wang, Hongshuai and Lin, Haitao and Xu, Hongteng and Gao, Zhifeng},
  booktitle={International Conference on Machine Learning},
  pages={63002--63015},
  year={2025},
  organization={PMLR}
}