/deepDTnet

Primary LanguageMATLAB

deepDTnet

paper "Target Identification among Known Drugs by Deep Learning from Heterogeneous Networks"

Quick start

we provide an example script to run experiments with the features already extracted from our dataset:

  • Run run_deepDTnet.m: predict drug-target interactions, and evaluate the results with cross-validation. Note: See the "Tutorial" section below for a detailed instruction on how to run deepDTnet on your own dataset.

Code and data

src/ directory

  • auc.m: evaluation script
  • compute_similarity.m: compute Jaccard similarity based on association network(i.e., drug-disease, drug-side-effect, and protein-disease networks)
  • imf_train.mexa64: pre-built binary file of PU-Matrix Completion algorithm
  • PUMC.m: predict drug-target interactions (DTIs) with PU-Matrix Completion
  • run_deepDTnet.m: example code of running deepDTnet for drug-target prediction

data/ directory

  • drug_dict.txt: list of drug unique identifier and drug names
  • protein_dict.txt: list of protein unique identifier and protein names
  • disease_dict.txt: list of disease unique identifier and disease names
  • se_dict.txt: list of side effect unique identifier and side effect names
  • drugdrug.txt: Drug-Drug interaction matrix
  • drugDisease.txt: Drug-Disease association matrix
  • drugsideEffect.txt: Drug-SideEffect association matrix
  • drugsim1network.txt: Drug chemical similarity matrix
  • drugsim2network.txt: Drug therapeutic similarity matrix
  • drugsim3network.txt: Drug sequence similarity matrix
  • drugsim4network.txt: Drug biological processes similarity matrix
  • drugsim5network.txt: Drug cellular component similarity matrix
  • drugsim6network.txt: Drug molecular function similarity matrix
  • proteinprotein.txt: Protein-Protein interaction matrix
  • proteinDisease.txt: Protein-Disease association matrix
  • proteinsim1network.txt: Protein sequence similarity matrix
  • proteinsim2network.txt: Protein biological processes similarity matrix
  • proteinsim3network.txt: Protein cellular component similarity matrix
  • proteinsim4network.txt: Protein molecular function similarity matrix

feature/ directory

We provided the pre-trained vector representations for the 15 networks, which were used to produce the results in our paper.

DNGR/ directory

This directory contains code necessary to run the DNGR algorithm.

  • Run main.m: an example script to generate a low-dimensional vector representation of features for each node in the Drug chemical similarity network. Note: For different networks, you need to fine tune the parameters in BuildNets.m. The original version of this code base was originally from http://shelson.top/, and we owe many thanks to Shaosheng Cao for making his code available.

supplementary/ directory

  • Supplementary_Table_3.xlsx: an excel table including 4,978 experimentally validated drug-target interactions connecting 732 approved drugs and 1,915 human targets.
  • Supplementary_Table_4.xlsx: an excel table including all target lists for G-protein-coupled receptors (GPCRs), kinases, nuclear receptors (NRs), and ion channels (ICs).
  • Supplementary_Table_5.xlsx: an excel table including novel drug-target interactions predicted by deepDTnet with score above the threshold.

Tutorial

We test the code on Ubuntu 16.04 with Matlab R2016a installed

  1. Put interaction/association/similarity matrices in the data/ folder.
  2. Create a simNet/ folder under deepDTnet/ and run compute_similarity.m in the src/ folder, which will compute the Jaccard similarity of drugs and proteins, based on the three association matrices mentioned above.
  3. run main.m in the DNGR/core folder for 2 interaction networks ,10 similarity networks in the data/ folder and 3 similarity networks in the simNet/ folder, which will learn the feature vectors of drugs and proteins for the 15 networks and save them in the feature/ folder.
  4. Specify parameters and set the path of feature vectors in run_deepDTnet.m and execute it. This script will predict the drug-target interactions and evaluate the results using a five-fold cross-validation.