Research On Disease-ncRNA Association Model Based On Multidimensional Biomolecular Network

With the development of the ‘Next-generation' sequencing technology(highthroughput genome sequencing),more and more non-coding RNA(ncRNA)molecules have been discovered.These ncRNAs play important roles in various biological processes,and their mutation or disorder may be the incentive of human diseases.It is of great significance to systematically study the relationships between ncRNA and disease.However,owing to the massive amounts of biological information,it is not desirable in terms of time efficiency and cost to trace the information of pathogenic ncRNAs through traditional biological experiments.Therefore,it is one of the hotspots in the field of bioinformatics to develop efficient methods to discover pathogenic ncRNA-related informations.In this paper,we propose several models to infer the potential relationships between disease and ncRNA.Based on the network model,LDLMD was proposed to predict the associations between disease and single type ncRNA.Based on the recommendation system,PMFILDA was proposed to predict the association between disease and lncRNA.Then,based on LDLMD,a new prediction model PADLMP was proposed to infer the relationships between disease and lncRNA-miRNA pairs.In the LDLMD model,the lncRNA-miRNA-disease interaction network was constructed based on the known lncRNA-miRNA association,miRNA-disease correlation,and then we developed a new similarity computational model.Finally,the link-predictive method was used to identify potential disease-lncRNA associations.In PMFILDA,we integrated lncRNA-miRNA association network,miRNAdisease association network,and lncRNA-disease correlation network to construct a lncRNA-disease weighted association network.And then,we adopted the KNN algorithm based on the semantic similarity of diseases and the similarity of lncRNA functions to further update the weighted association network.Thereafter,the potential disease-lncRNA associations were inferred based on the probability matrix factorization.In the PADLMP model,we have improved the LDLMD method.We did not only add the known lncRNA-disease associations in PADLMP,but also predicted target has changed from a single type of pathogenic ncRNA to multiple types of pathogenic ncRNA.Finally,to evaluate the superiority of the new prediction models,we performed Leave One Out Cross-Validation(LOOCV)based on known related datas and the results show that LDLMD,PADLMP and PMFILDA achieve the AUC values of 0.8925,0.9318 and 0.8794 respectively.Comparing with other methods,our models can achieve better than them.Moreover,case studies were implemented to further estimate the performance of them,and simulation results illustrated that LDLMD,PADLMP and PMFILDA could achieve satisfying prediction performance as well.