deepDR

deepDR: A network-based deep learning approach to in silico drug repositioning

https://doi.org/10.1093/bioinformatics/btz418

https://github.com/ChengF-Lab/deepDR

dataset directory

Contain the drug-disease interactions dataset.

PPMI directory

Contain the PPMI matrices of ten drug-related networks.

models directory

1. deep_network_fusion directory, which contains 3 variants implementations of Multimodal Deep Autoencoder (MDA)

   https://doi.org/10.1093/bioinformatics/bty440

2. recommendation directory, which contains implementation of Collective Variational Autoencoder (cVAE)

   https://doi.org/10.1145/3270323.3270326

   • Updated for calculating Recall@K metric.

Tutorial

1. To get drug features learned by MDA, run

   python get_features.py params.txt
   

2. To predict drug-disease associations by cVAE, run
   1. Pretraining with features:

      python models/recommendation/collective_variational_autoencoder.py --dir dataset -a 6 -b 0.1 -m 300 --save --layer 1000 100
      

   2. Refine training with rating:

      python models/recommendation/collective_variational_autoencoder.py --dir dataset --rating -a 15 -b 3 -m 500 --load 1 --layer 1000 100
      

• On colab, simply follow the deepDR.ipynb notebook.