MINDG: A Drug-Target Interaction Prediction Method Based on an Integrated Learning Algorithm

This papaer has been received by Bioinformatics 2024. This repository contains script which were used to build and train the MINDG model together with the scripts for evaluating the model's performance.

Dependency

python                  3.6.8
torch                   1.5.0
numpy                   1.19.1
scipy                   1.5.0
torch_sparse            0.6.4
pandas                  1.1.5
scikit-learn            0.22.1
matplotlib              3.2.2
requests                2.27.1
pandas-flavor           0.6.0
subword-nmt             0.3.8
prettytable             0.7.2
texttable               1.7.0
tqdm                    4.65.0
pytdc                   0.4.1
loguru                  0.7.2

Dataset

The dataset used in the experiments are provided as follows:

BindingDB dataset is hosted in https://www.bindingdb.org/bind/index.jsp.

Davis Dataset can be found in http://staff.cs.utu.fi/~aatapa/data/DrugTarget/.

The BingDB and DAVIS datasets are automatically downloaded from TDC by Main.py. If you want some raw csv samples, please click BindingDB, DAVIS to download.

Train and Test

Train

Navigate to the project source directory

cd src

Config the training parameters by adjusting the input parameters of run function

run('DAVIS', phase="train",batch_size=32,epochs=5,learning_rate=5e-4,lr_step_size=10,seed_id=10,device=torch.device('cpu'))

Description of run function parameters

name            dataset name, "BingdingDB_kd" or "DAVIS"
phase           work phase, "train"/"test"
batch_size      batch size of data ,default 32
epochs          number of train epoch, default 5
learning_rate   learning rate, default 5e-4
lr_step_size    Scheduling step size for learning rate, default 10
seed_id         random seed id, default 10

start training

python Main.py

The directory to save Model parameters is output/model. Model file name is model name + dataset name + epoch number.pt

(mindg) yang@yang:~/sda/github/MINDG/output/model$ ll
total 119032
drwxrwxr-x 2 yang yang     4096 Jan  9 21:16 ./
drwxrwxr-x 6 yang yang     4096 Dec 29 19:58 ../
-rw-rw-r-- 1 yang yang 16303055 Jan  3 10:33 mindg_BindingDB_Kd_epoch10.pt
-rw-rw-r-- 1 yang yang 18788185 Jan  4 08:38 mindg_BindingDB_Kd_epoch20.pt
-rw-rw-r-- 1 yang yang 18788303 Jan  9 15:34 mindg_BindingDB_Kd_epoch5.pt
-rw-rw-r-- 1 yang yang 13408711 Dec 26 06:09 mindg_DAVIS_epoch100.pt
-rw-rw-r-- 1 yang yang 13408593 Dec 30 17:13 mindg_DAVIS_epoch20.pt
-rw-rw-r-- 1 yang yang 13408711 Jan  9 11:36 mindg_DAVIS_epoch5.pt
-rw-rw-r-- 1 yang yang 13408593 Dec 28 18:52 mindg_DAVIS_epoch90.pt
-rw-rw-r-- 1 yang yang 14355801 Jan  3 22:20 mindg_KIBA_epoch20.pt

Test

when phase is "test", The routine will load the Model file stored in the output directory.

run('DAVIS', phase="test",batch_size=32,epochs=5,learning_rate=5e-4,lr_step_size=10,seed_id=10,device=torch.device('cpu'))

start testing

python Main.py

There are two pre-trained models(mindg_BindingDB_Kd_epoch10.pt, mindg_DAVIS_epoch5.pt) for downloading. If you would like to obtain more pretrained models, please feel free to email yang hailong.

BibTeX

cite my paper

@article{10.1093/bioinformatics/btae147,
    author = {Yang, Hailong and Chen, Yue and Zuo, Yun and Deng, Zhaohong and Pan, Xiaoyong and Shen, Hong-Bin and Choi, Kup-Sze and Yu, Dong-Jun},
    title = "{MINDG: A Drug-Target Interaction Prediction Method Based on an Integrated Learning Algorithm}",
    journal = {Bioinformatics},
    pages = {btae147},
    year = {2024},
    month = {03},
    issn = {1367-4811},
    doi = {10.1093/bioinformatics/btae147},
    url = {https://doi.org/10.1093/bioinformatics/btae147},
    eprint = {https://academic.oup.com/bioinformatics/advance-article-pdf/doi/10.1093/bioinformatics/btae147/56970399/btae147.pdf},
}

Acknowledgement

The code is based on HOGCN and DeepPurpose.