The Study Of Drug Target Prediction Based On Network Pharmacology Space

The first key to modern drug discovery is to find,identify and prepare the drug molecule targets.Especially,in the drug development process,several types of proteins such as enzymes,ion channels,G-protein coupled receptor(GPCR)and nuclear receptors represent the vast majority of current drug targets.However,due to the influence of the throughput,accuracy and cost,it is difficult to carry out extensively by using the conventional experimental methods to illustrate the using of these potential interactions of drug-target.There is an urgent need to develop effective methods to help researchers to mine the law of the interaction between drugs and targets,thus providing complementary and supplementary evidence for experiments.Aiming at these problems,by collecting reliable data sources,integrating a variety of biological information,such as protein sequence features,compound efficacy similarity,structural similarity compounds and drug target features such as network topology,we can predict the potential drug-target effectively by using the methods of the supervised learning.The main content can be outlined as follows:1.Some researchers who use distance learning and regression model to predict the drug-protein interactions in the previous studies.In order to resolve the problem of the prediction of the drug-protein interaction,we try to use shared degree rather than the shortest distance and linear regression models to integrate drug space and gene space to pharmacological space.In addition,we have fused efficacy space to the pharmacological space for prediction,which make the prediction more accurately and comprehensively.According to the known four protein family(enzyme,GPCR,ion channels and the nuclear receptors)and the drug data of the interaction with it,we propose a method of supervised learning based on modified bipartite graph to predict drug-target interaction network.We use the drug-target interaction network topology information and integrate a variety of biological characteristics to predict potential the relationship of the drug-target.The results show that the proposed method has better performance than the existing bipartite graph model method.2.The number of the labeled samples for supervised learning is less and the number of unlabeled samples is larger,which make it difficult to obtain the generalization ability of the model.So,if we can use the larger unlabeled data to improve the generalization ability of the model on the less labeled data,this is a major concern in our study.From the point of the chemical,drug efficacy spaces,geneticspace,starting at four known interaction network of drug-target to maximize the use of existing data to predict a large number of unlabeled date,we present a improved semi-supervised learning methods to mine the potential drug-target from the known network of the drug-target interactions.Experimental results show that the performance of the proposed method was superior to the existing methods.It not only improves the prediction accuracy but also effectively predict a number of potential drug-target relationships which has been validated in the authority of the relevant database.