The Research On Drug Repositioning Based On Network Structure Analysis

In order to understand the pathogenesis of human beings and improve the scientific and comprehensive of drug treatment,the relationship between diseases and drugs has been a popular problem and research focus of modern medicine.In the past few decades,with the progress of medical level,the research on the therapeutic effect and mechanism of drugs has gradually established a complex pharmacology research system,among which the relationship between drugs and targets has become the focus and difficulty of drug research.Because the research and development of new drugs needs to bear high time and economic costs,the research on new treatment approaches of approved drugs has become a popular research problem,which is called drug repositioning.In the research of drug repositioning,understanding and integrating the overall relationship among various biomedical entities is still a huge challenge,which mainly refers to the relationship of drug-drug,drug-target and target-disease which are closely related to drug research and development.From the perspective of bioinformatics,this paper researches the related problems between drug and disease relying on the existing computer technology and methods.Mainly by exploring the relationships between drug,target and disease,this paper predicts drug-target interactions and functional modules of protein-protein interactions.After understanding and analyzing the current research status and methods of above problems,an integrated model is constructed based on the relevant biomolecular network data,and several methods are proposed for experimental calculation and validation.The innovation points of the paper are as follows:(1)According to the research on drug-target interaction and many heterogeneous multi-source databases,it can be seen that traditional biochemical methods are difficult to screen some problems on a large scale.Therefore,this paper proposes a regularized least square method of Kronecker product based on the triple heterogeneous network model of drugs,targets and diseases(THN_KRLS),in order to better predict drug-target interaction.This method constructs the similarity matrix space via various similarity calculation methods,predicts the matrix by using the regularized least square method of Kronecker product,and finally verifies the prediction results through cross validation and external database.The results show that the method is superior to FLapRLS and RLS-Kron methods,and the interaction of higher-ranked drug targets has been proved by other authoritative research and experiments.(2)In view of the poor prediction results of new drugs and targets,this paper proposes a cascade deep forest method based on the triple heterogeneous network model of drugs,targets and diseases.This method uses multi-granularity scanning to process multiple similarity matrix as feature input,and multiple random forest integration classifiers with cascade characteristics for training.Each layer takes the training results as the input of the next layer,and the final prediction results are obtained when they cannot be significantly improved.The experiment shows that the method is superior to RLS-KF,RF,DTiGEMS and iDEI-ESBoost.(3)The actual target network is mostly heterogeneous,multi-sourced and interconnected.Hence a method of modular decomposition for target layer is proposed in this paper.This method can describe the global topology of target network quickly and losslessly and distinguish the interaction within the module in detail.The results show that the method can help to analyze the interaction and working principle of target protein in biological system in large-scale network structure,and further deepen the research on the mechanism and response of target protein in specific physiological state.This method is of great significance to the analysis of protein function principle.