Home

Awesome

DTINet: A Network Integration Approach for Drug-Target Interaction Prediction

:loudspeaker::collision::new: Updates: A Python implementation of DTINet can be found in this GitHub repo: PyDTINet.

DTINet is a computational pipeline to predict novel drug-target interactions (DTIs) from heterogeneous network. DTINet focuses on learning a low-dimensional vector representation of features for each node in the heterogeneous network, and then predicts the likelihood of a new DTI based on these representations via a vector space projection scheme. See our paper on Nature Communications.

Quick start

We provide an example script to run experiments on our dataset:

Note: See the "Tutorial" section below for a detailed instruction on how to specify parameters of DTINet, or how to run DTINet on your own dataset.

Code and data

src/ directory

data/ directory

feature/ directory

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

Third-party software

Our implementation requires the Inductive Matrix Completion (IMC) library. We provided an executable binary file (src/train_mf.mexa64) for convenience. The executable binary file was built on a typical Ubuntu 14.04 (64 bit) system.

We also provided the source code of IMC (src/leml-imf-src.zip) since the downloads from the IMC website is currently unavailable. If you are using other Linux platforms or have issues using the provided binary file, please consider building the library from source code by running bash install_imc.sh in the src/ folder.

Tutorial

  1. Put interaction/association matrices in the data/ folder.
  2. Create a network/ folder under DTINet/ and run compute_similarity.m, which will compute the Jaccard similarity of drugs and proteins, based on interaction/association matrices.
  3. Specify parameters (number of dimensions of feature vectors, restart probability, the maximum number of iterations) and run run_DCA.m, which will learn the feature vectors of drugs and proteins and save them in the feature/ folder.
  4. Set the path of feature vectors and corresponding parameters in run_DTINet.m and execute it. This script will predict the drug-target interactions and evaluate the results using a ten-fold cross-validation.

Supplementary Information

supplementary/ directory

Citation

Luo, Y., Zhao, X., Zhou, J., Yang, J., Zhang, Y., Kuang, W., Peng, J., Chen, L. & Zeng, J. A network integration approach for drug-target interaction prediction and computational drug repositioning from heterogeneous information. Nature Communications 8, (2017).

@article{Luo2017,
  author = {Yunan Luo and Xinbin Zhao and Jingtian Zhou and Jinglin Yang and Yanqing Zhang and Wenhua Kuang and Jian Peng and Ligong Chen and Jianyang Zeng},
  title = {A network integration approach for drug-target interaction prediction and computational drug repositioning from heterogeneous information},
  doi = {10.1038/s41467-017-00680-8},
  url = {https://doi.org/10.1038/s41467-017-00680-8},
  year  = {2017},
  month = {sep},
  publisher = {Springer Nature},
  volume = {8},
  number = {1},
  journal = {Nature Communications}
}

Contacts

If you have any questions or comments, please feel free to email Yunan Luo (luoyunan[at]gmail[dot]com).