PocketAnchor

Learning Structure-based Pocket Representations for Protein-Ligand Interaction Prediction.

Please refer to our Cell Systems paper for more detailed information.

Cite our paper by

@article{li2022pocketanchor,
  title={PocketAnchor: Learning Structure-based Pocket Representations for Protein-Ligand Interaction Prediction},
  author={Li, Shuya and Tian, Tingzhong and Zhang, Ziting and Zou, Ziheng and Zhao, Dan and Zeng, Jianyang},
  journal={Cell Systems},
  volume={14},
  number={8},
  pages={692-705.e6},
  year={2023},
  issn={2405-4712},
  doi={https://doi.org/10.1016/j.cels.2023.05.005},
  url={https://www.sciencedirect.com/science/article/pii/S2405471223001497},
}

The code for data processing can be found in https://github.com/lishuya17/PocketAnchorData.

The processed data can be found in docker image: https://hub.docker.com/r/tiantz17/pocketanchor-models. (Not recommended)

(Update) You can pull another docker image containing code, data, environment, trained models, and prediction results for reproduction: https://hub.docker.com/r/tiantz17/pocketanchor.

1. Requirements

cuda                11.2
python              3.7.4
torch               1.7.1
torch-geometric     1.6.3
numpy               1.19.0
pandas              1.2.4
rdkit               2020.03.3.0
scikit-learn        0.21.3 
scipy               1.6.3 
tensorboard         2.4.1

2. Reproducing results

1. Prepare a environment that satisfying the above requirements;
2. Download the trained model and the input data files in docker image;
3. Extract the following files and unzip into this folder;
   • PocketAnchor-models.zip
   • PocketAnchor-data-Affinity.zip
   • PocketAnchor-data-PocketDetection.zip
4. Run the inference scripts below (run time ranges from a few minutes to a couple of hours depending on the size of dataset);
5. The results can be found in [TASK]/results/[FOLDER]/.

1. PocketAnchor-site

Protein ligand binding site prediction

python runPrediction.py --task PocketDetection --dataset COACH420
python runPrediction.py --task PocketDetection --dataset HOLO4k

2. PocketAnchor-affinity

Protein-ligand binding affinity prediction

python runPrediction.py --task Affinity --dataset CASF --setting original --info original
python runPrediction.py --task Affinity --dataset CASF --setting newprotein --info newprotein
python runPrediction.py --task Affinity --dataset CASF --setting expanded --info expanded

3. Train PocketAnchor

1. Prepare a environment that satisfying the above requirements;
2. Generate anchor positions and the corresponding features of customized dataset following PocketAnchorData.
3. Run the training scripts below;
4. The trained models can be found in [TASK]/models/[FOLDER].

1. PocketAnchor-site

Protein ligand binding site prediction

python runTrain.py --task PocketDetection --dataset scPDB

2. PocketAnchor-affinity

Protein-ligand binding affinity prediction

python runTrain.py --task Affinity --dataset CASF --setting original --info original
python runTrain.py --task Affinity --dataset CASF --setting newprotein --info newprotein