HDN-DDI: drug-drug interactions prediction with hierarchical molecular graphs and enhanced dual-view representation learning

Authors: Jinchen Sun, Haoran Zheng*

Article Link: None (the url will be given after the paper is accepted.)

Note that: The relevant figures will be published in this repository after the paper is published. Please wait patiently.

Requirement

To run the code, you need the following dependencies:

PyTorch >= 1.9.0
PyTorch Geometry == 2.0.3`
rdkit == 2020.09.2

Reimplement

The average performace of HDN-DDI can be directly calculated by evaluate.ipynb.

If you have configured the environment and want to verify it yourself:

1. Download the DrugBank and Twosides (Note: Twosides requires unzipping the 7z files), and place the folder according to the following requirements:

- drugbank_test /
    - DrugBank /
        - cold_start / ...
        - warm_start / ...
        - ddis.csv
        - drug_smiles.csv
        - id_data_dict_dsn_full_connect.pkl
    - transductive_test.py
    - inductive_test.py
    - ...

- twosides_test /
    - Twosides /
        - fold0 / ...
        - fold1 / ...
        - fold2 / ...
        - ddis.csv
        - drug_smiles.csv
        - id_data_dict_dsn_full_connect.pkl
    - test.py
    - ...

2. Run the script to complete multiple experiments (Note: You can modify the script's comment variable to customize the comments for each experiment):

./repeat.sh

3. Calculate the average performance of the model through evaluate.ipynb