Prediction Of MiRNA-disease Association Based On Matrix Completion And Bi-random Walk

MicroRNAs(miRNAs)are non-codingRNA molecule with length of about 21 nucleotides.Controlling the degradation and expression of mRNA through base pairing.Studies have shown that miRNA plays an essential role in a variety of biological processes necessary for humans,such as cell proliferation,differentiation,death and so on.Lot of experimental data and case studies show that miRNA is closely related to human complex diseases.MiRNA mutations and abnormal regulation are associated with many diseases,so,identifying the association between miRNAs and diseases can help explain the pathogenesis of complex human diseases,and can help diseases treatment and prevent occurrence of diseases.Therefore,it is particularly important to use existing relevant data to predict the association between miRNAs and diseases.Based on a previous research work,in this thesis we propose a novel method MCBRWMDA for predicting miRNA-disease association based on matrix completion and bi-random walk.First,we collect and sort out the data about miRNAs and diseases in the database HMDDV2.0.After sorting and screening,495 miRNAs,385 diseases and 5430 verified associations are obtained.Second,we calculate miRNA’s functional similarity,disease’s semantic similarity,and the nuclear similarity of Gaussian interaction profiles.Third,we use the matrix completion algorithm to reconstruct a new matrix and integrate the existing matrix to get the final similarity matrix between miRNA and disease.Last,we preprocess the known association matrix to obtain a new association network using weighted K nearest neighboring algorithm,using the obtained three networks,we construct a binary heterogeneous network to perform bi-random walks to predict potential miRNA-disease associations.Experimental results show that the MCBRWMDA model achieves reliable prediction performance.The global leave-one-out cross-validation AUC value is 0.9733,and the five-fold cross-validation AUC value is 0.9412.Five-fold cross-validation is performed on a single common disease,and the average value of AUC for 20 diseases can achieve 0.9375.In the case analysis of breast cancer,esophageal cancer and lung cancer,the miRNAs predicted for these diseases are sorted according to their scores from high to low,and the top 50 miRNAs are verified one by one in the latest database HMDDV3.2 and db DEMC2.0.The results obtained show 100%,88% and 98% of breast cancer,esophageal cancer and lung cancer of the top 50 prediction results were successfully verified.These results prove the validity and efficiency of the MCBRWMDA model.