Skip to content

Latest commit

 

History

History
60 lines (45 loc) · 2.41 KB

README.md

File metadata and controls

60 lines (45 loc) · 2.41 KB
Raw

S-MAN: Spatial-aware Molecule Graph Attention Network

中文版本 English Version

DiStance-aware Molecule Graph Attention Network (S-MAN) is a novel deep learning framework to predict drug-target binding affinity(DTA). This is the implementation of S-MAN based on PaddlePaddle and PGL. The paper about this implementation is: Distance-aware Molecule Graph Attention Network for Drug-Target Binding Affinity Prediction.

Datasets

The PDBbind dataset can be downloaded here. After downloading the data, you should first preprocess dataset to generate the protein-ligand graph and features. The preprocessed Protein-Ligand graph and feature can be downloaded here: Protein-Ligand dataset.

You can also run the following commnd to preprocess the PDBbind dataset to generate the protein-ligand graph and features.

python preprocess.py --data_path YOUR_DATASET_PATH --dataset_name v2016_LPHIN3f5t_Sp --output_path YOUR_OUTPUT_PATH --cutoff 5

PS: cutoff is the threshold of distance cutoff between atoms.

Dependencies

  • networkx >= 2.1
  • paddlepaddle >= 1.8.4
  • pgl >= 1.1.0
  • openbabel == 3.1.1 (optional)

How to run

For examples, use gpu to train S-MAN on PDBbind dataset.

python train.py --lr_d --data_path YOUR_DATA_PATH --dataset v2016_LPHIN3f5t_Sp --save_path MODEL_SAVE_PATH --gpu YOUR_DEVICE

You can also test the saved models as follows:

python test.py --data_path YOUR_DATA_PATH --dataset v2016_LPHIN3f5t_Sp --model_path YOUR_MODEL_PATH --gpu YOUR_DEVICE

Hyperparameters

  • dataset: name of dataset

  • num_layers: number of GNN layers

  • dist_dim: dimensions or buckets of distance spliting

  • lr: learning rate

  • lr_d: use learning rate decay

  • drop: dropout ratio

  • data_path: file path of dataset

  • save_path: the path to save model （e.g, ./runs)

  • model_path: file path of the saved model (e.g, ./runs/SMAN)

Reference

S-MAN

@article{zhou2020distance, title={Distance-aware Molecule Graph Attention Network for Drug-Target Binding Affinity Prediction}, author={Zhou, Jingbo and Li, Shuangli and Huang, Liang and Xiong, Haoyi and Wang, Fan and Xu, Tong and Xiong, Hui and Dou, Dejing}, journal={arXiv preprint arXiv:2012.09624}, year={2020}, url={https://arxiv.org/abs/2012.09624} }