Disease-gene-drug Association Prediction Based On Multiplex Network Algorithms

Cancer is a type of malignant tumor that seriously affects human life and health.Studies have shown that the occurrence of complex diseases such as cancer is often related to the mutation of key genes in human body.In cancer treatment,anticancer drugs can act on key cancer genes to induce apoptosis of cancer cells or restore mutated genes to normal.Now many anti-cancer drugs have appeared on the market,but that is far from enough.Meanwhile the research cost of new drugs is very expensive and the development cycle is very long,which seriously affects the progress of cancer treatment.Therefore this paper mainly studies the existing anticancer drugs by integrating disease phenotype,gene expression,and drug response data from different databases,using multiplex network-based algorithms to study the relationship between drugs,genes and diseases.Gene-drug association predictions obtained from experimental results can provide reference value for other cancer patients with the same pathogenic gene during treatment.And drug repositioning results obtained from experiments can provide theoretical basis for the treatment of other complex diseases.The work of this paper is as follows:(1)On the premise that similar drugs may have effects on similar genes,gene expression and drug response data of lung cancer are obtained from GDSC database,and gene expression similarity network and drug response similarity network are constructed respectively using Pearson correlation coefficient.Known gene-drug associations link two networks to form a dual heterogeneous network.On this basis,this paper proposes the multiplex network algorithm based on graph entropy(MNA＿GE),which identifies gene-drug module through three main steps in gene-drug network.MNA＿GE algorithm identifies five gene-drug modules in the dual heterogeneous network.Through experiments,it is found that these modules have obvious biological significance in GO enrichment analysis and KEGG enrichment analysis,and some of identified gene-drug relationship has been verified by the literature search method.When compared with other methods,MNA＿GE algorithm shows higher module discrimination accuracy in simulated datasets with different noise levels and different proportions of irrelevant samples.(2)On the basis of using dual heterogeneous network to predict gene-drug association,this paper further studies the complex interaction between multiple diseases and multiple drugs on disease-gene-drug multiplex network,so as to realize disease-drug association prediction.Multi-omics data related to diseases,genes,and drugs are first obtained from GDSC,CCLE,OMIM and Drug Bank databases.Disease phenotype similarity network,gene expression similarity network and drug response similarity network are constructed respectively.Then gene expression similarity network is used as an intermediate layer,and three similarity networks are integrated into a multiplex heterogeneous network framework using known disease-gene associations and gene-drug associations.On this basis,this paper proposes the drug repositioning algorithm based on multiplex network(DRMN).First,DRMN uses Page Rank algorithm to calculate PR value of each node in the multiplex network,then calculates and sorts the association score between each disease-drug pair,and finally achieves the purpose of disease-drug association prediction.Applying DRMN algorithm to two datasets,among top 10%,30%,and 50% of obtained association prediction scores,there has been multiple pairs of association predictions that have been verified as effective disease-drug combinations by existing results.In comparison with other algorithms,DRMN algorithm has better performance in many evaluation indicators including the AUC value.