Skip to content

[ICML 2024] Interaction-based Retrieval-augmented Diffusion Models for Protein-specific 3D Molecule Generation

License

Notifications You must be signed in to change notification settings

YangLing0818/IRDiff

Repository files navigation

Interaction-based Retrieval-augmented Diffusion Models for Protein-specific 3D Molecule Generation (ICML 2024)

Official implementation for our ICML 2024 paper - Interaction-based Retrieval-augmented Diffusion Models for Protein-specific 3D Molecule Generation.

Alt text

Environment

conda env create -f irdiff.yaml
conda activate irdiff

Data and Preparation

The data preparation follows IPDiff. For more details, please refer to the repository of TargetDiff.

❗️❗️Path to the pre-trained PMINet and pre-trained IRDiff:

./pretrained_models

❗️❗️Path to the indices of retrieved prompts:

./src

Training

conda activate irdiff
python train.py

Sampling

python sample_split.py --start_index 0 --end_index 99 --batch_size 25

Evaluation

python eval_split.py --eval_start_index 0 --eval_end_index 99

Calculate metrics

python cal_metrics_from_pt.py

Results

Alt text

Citation

@inproceedings{
    huang2024interactionbased,
    title={Interaction-based Retrieval-augmented Diffusion Models for Protein-specific 3D Molecule Generation},
    author={Zhilin Huang and Ling Yang and Xiangxin Zhou and Chujun Qin and Yijie Yu and Xiawu Zheng and Zikun Zhou and Wentao Zhang and Yu Wang and Wenming Yang},
    booktitle={International Conference on Machine Learning},
    year={2024},
}