Drug-drug interaction (DDI)

The proposed model consists of a graph convolutional autoEncoder network (GCAN) for embedding drug induced transcriptome data from the L1000 database of the Integrated Network-based Cellular Signatures (LINCS) project; and a long short-term memory network (LSTM) for DDI prediction. For a case study, we applied the proposed deep-learning model to antidiabetic agents.

File directory structure

"/data/":          folder of experiment data

"/data_preprocessing/GCAN/": use the GCAN framework for data preprocessing

"/model_train/GCAN_LSTM/" :  the GCAN + LSTM framework code

"/compare/":          six types of comparison model code

"/result/":           folder of results

The required toolkit support and verified version number for project execution

toolkit	numpy	pandas	tensorflow	skmultilearn	sklearn
version number	1.16.3	1.0.3	1.13.1	0.2.0	0.21.2

Code instructions

1. data preprocessing

   inits.py Layer.py Model.py : initialization of the GCAN framework

   train.py : the GCAN framework is used for data preprocessing

2. model training

   train_GEDDI_model.py param1 param2 : use GCAN + LSTM framework to train model. param1: train data; param2: test data

3. Compare the model

   compare_model.py param1 param2 : contrast models include "MLARAM", "MLkNN", "BRkNNa", "BRkNNb", "RF", "MLTSVM". param1: train data; param2: test data