NRWRH: Network-based Random Walk with Restart on the Heterogeneous network

Reference#

• Xing Chen, Ming-Xi Liu and Gui-Ying Yan. Drug-Target Interaction Prediction by Random Walk on the Heterogeneous Network. Submitted to PLoS ONE.

Method#

The aim of NRWRH is to predict the potential target proteins of the given drug. NRWRH is proposed based on the assumption that similar drugs often target similar target proteins and the framework of Random Walk with Restart (RWR). The method is composed of four steps: firstly, three networks (protein-protein similarity network, drug-drug similarity network, and known drug-target interaction network) are constructed and combined into a heterogeneous network by known drug-target interactions; secondly, the initial probability of random walk is determined to make random walk start at the given drug nodes and seed target nodes simultaneously; then random walk on the heterogeneous network is implemented; finally the most probable targets are selected according to the stable probability of the walk. To our knowledge, there are no computational methods to use the idea of random walk to predict potential drug-target interactions before our paper. NRWRH is different from traditional random walk with restart in two aspects. The first is that the information of known drug-target interaction network is integrated to improve drug chemical structure similarity and protein sequence similarity. The other difference is that the random walk is implemented on three networks. When searching for targets of a drug which having known targets in the network, candidate targets can be ranked by calculating the similarity between candidate targets and known targets. However, if the drug has no known target, only using target similarity will be insufficient and hence drug similarity must be used. In this case, potential targets of this given drug were selected based on target information of drugs which are similar to the given drug.

Procedure#

nrwrhdrugtarget.m is used to give the prediction about the potential targets of the given drug by Network-based Random Walk with Restart on the Heterogeneous network. It can be run by MATLAB. The input variable "backprobability" means that the random walker can return to seed nodes with this probability in each step. Variable “lanbuda” is the jumping probability. Large lanbuda introduces more mutual dependency of ranking between drugs and targets. Parameter “yita” controls the impact of two kinds of seed nodes. Parameter “gamad1”, “gamad2”, “gamat1”, “gamat2” are all the linear coefficient of integrating the known drug-target interactions to improve the drug-drug similarity and target-target similarity. Parameter “drugID” means you want to predict potential targets of this drug. The out variable “top” means the final ranking of candidate targets for the given drug. Before the usage of the program, you should prepare several files: (1) chemicalsimilarity.txt showing the similarity of all the drugs in the network; (2) sequencesimilarity.txt showing the similarity of all the targets in the network; (3) interaction.txt showing the known drug-targets interactions in the network. You can read txt files of the example for reference of the format. If there are some problems in the process of usage of the code, please be free to contact me: xingchen@amss.ac.cn.

Code#

• nrwrhdrugtarget.m
• combinedsimilarity.m
• example.rar