StructuralDPPIV

This project is made possible by the Wei Group @ Shandong University. All the code is under MIT License.

Project code of the paper StructuralDPPIV: A novel atom-structure based model for discovering dipeptidyl peptidase-IV inhibitory peptides.

Installation

Requirements

dependencies:
  - Python 3.10.13 (recommonded)
  - torchaudio
  - torchvision
  - pytorch
  - cudatoolkit=11.3
  - pyg
  - rdkit
  - pytdc
  - pytorch-lightning==1.9
  - nb_conda
  - transformers
  - pynvml
  - grad-cam
  - tdqm

Usage

Training

For training the model, you can run main/train.py to train the model. The training process will be saved in the ./logs directory. To see the training process, you can run tensorboard --logdir ./logs and open the browser to http://localhost:6006/.

Testing

We provide a trained model in https://drive.google.com/drive/folders/18Kb81AED8_dNEdQ_uQ8jYtyoPmOySd_a?usp=drive_link. You can download the model and put it in the ./ckpt directory. Then you can run main/test.py to test the model, or use your own data/trained model to see the results.

Contacts

If you have any questions or suggestions, please contact us here, or mail to scholarwd [at] gmail [dot] com.