The Research On Drug Repositioning Algorithm By Multiple Information Integration

Drug Repositioning,which is to find new indications for existing drugs,can effectively shorten the the development cycle of new drugs,reduce costs,and avoid risks.And it is becoming an important strategy for drug development.With the accumulation of related research in recent years and the open use of drug databases,Drug Repositioning based on computational analysis has become an important research direction.On the one hand,abnormal expression of genes cases the occurrence and development of some diseases.Drug molecules achieve therapeutic effects by combining with targets(such as proteins)in the body.Researchers can integrate drug-target and disease-target association data to explore more effective potential drug-disease associtations;On the other hand,the known relationship between drugs and diseases is relatively sparse.How to obtain higher drug repositioning accuracy has also come to the foreground.In this paper,based on the known drug-disease associtations,we integrate target information and propose a drug repositioning prediction algorithm based on drug-disease-target heterogeneous network.The main research of this paper is as follows:An algorithm based on Information Propagate in Heterogeneous Networks(IPHN)is proposed to explore the relationship between drugs and diseases.A heterogeneous network is constructed using known information of drugs,diseases and target.Then a Meta-Path based Approach is adopted to calculate the similarity scores for different types of nodes(drugs,diseases,and targets)and the scores are used to update the weights of network edge.The transfer probability matrix of random walk is constructed.Two paths of information propagation are the basis,which are from the drug network to the target network and then to the disease network,and from the drug network to the disease network directly.Then the combination of two random walks,walking in networks and walking between networks,is made to achieve the spread of drug association probability score on heterogeneous networks.Sorting all the drugs according the probability score and recommending drugs for a given disease.According to the results of 10-fold cross-validation,the proposed IPHN algorithm in this paper is superior to the comparison algorithm in the AUC and AUPR values.The results of the case analysis also show the effectiveness of the IPHN algorithm.In order to solve the influence of the sparse characteristics of drug-disease associtations on the performance of the algorithm,drug repositioning algorithm based on Graph Regularized Transductive Regression(GRTR)is proposed.The basic idea of the algorithm is to treat diseases as labels while treating drugs and targets as the objects.Based on the heterogeneous networks composed of drugs,diseases and targets,the statistical information on the disease distribution where unlabeled objects' neighbors could work is collected,which can be used to estimate the relationship between unlabeled objects and diseases.Using both unlabel objects and labeled objects as input,a graph regularized transductive regression model for sparse associtations is proposed while the correlation scores of potential drug-disease associtations are iteratively updated to find strong associtations between drugs and diseases.The experimental results show that GRTR achieves higher AUC and AUPR values compared with other comparison algorithms.In addition,the estimated results between drugs and targets can be obtained at the same time.The case studies of the drug-disease associtations and disease-target associtations also verify the effectiveness of GRTR.