This repository proposed a multi-modal deep auto-encoders based drug representation learning method for DDI prediction, named DDI-MDAE. Moreover, we improved DDI-MDAE by applying a specialized random forest classifier in the positive-unlabeled (PU) learning setting, the improved method is called DDI-MDAE (PU learning).
The input data is a combination of the original data, including five adjacency matrices. We take the adjacency matrix A = {A(1), A(2), ..., A(T)} as the input of our model, where T denotes the number of input feature networks.
The input data should be an undirected graph in which node IDs start from 1 to N (N is the number of nodes in the graph). Each line contains two node IDs of drug and feature repectively indicating an edge in the graph.
Text file sample
0 1
2 5
...
- Python 3.6.8 :: Anaconda, Inc.
- Tensorflow 1.14.0
When using this code, you need to clone this repo and load all the files in the folder into your running environment first. Then, you need to adjust the pu parameter in the following line of code in pu_main.py.
CV_num, method_num, alpha, beta, gama, mu, dimension, drug_size, removed_ratio, learning_rate, pu = [3, 4, 0.1, 2, 0.1, 1e-5, 128, 2367, 0, 0.001, 1]
Next, you need enter the root directory and run the following code:
cd src
python pu_main.py
When the parameter pu is 0, the file pu_main.py implements the DDI-MDAE method to predict potential interactions between drug-drug pairs, and when the parameter pu is 1, the file pu_main.py implements DDI-MDAE (PU-learning) method to predict potential interactions between drug-drug pairs.