Study On MiRNA With Positive And Unlabeled Learning Strategy And Matrix Completion

MicroRNA(miRNA)is a non-coding RNA with a length about 22 nucleotides,which is transcribed and processed from the endogenous hairpin structure of cells.miRNAs are often used as biomarkers for disease diagnosis.And researchers also use miRNAs as targets of drugs for treatment.Therefore,exploring more miRNA-disease associations can help to understand the pathogenesis of disease and can promote the diagnosis,prognosis and treatment of the disease.Computation-based prediction models can effectively predict miRNAs most possible to be related to disease,thereby reducing the experimental cost of discovering new miRNA-disease associations.The data utilized in is paper include miRNA-disease association data,disease semantic similarity,miRNA functional similarity,the integrated disease similarity and the integrated miRNA similarity.Two computational methods were proposed in this paper,named IMCMDA and PUMDA,to predict potential miRNA-disease associations.In IMCMDA,we take the integrated disease similarity and the integrated miRNA similarity as side information,and then an inductive matrix completion algorithm is utilized to predict potential miRNA-disease associations.Local Leave-One-Out Cross Validation(LOOCV),global LOOCV and 5-fold cross validation were implemented to evaluate the performance of IMCMDA.In the experimental part of IMCMDA,we use IMCMDA to carry out case studies on five diseases: colon cancer,kidney cancer,lymphoma,breast cancer and esophageal cancer.Among the top 50 predicted diseaserelated miRNAs,42,44,45,50 and 49 miRNAs are verified by the databases,respectively.In PUMDA,we first stitch the integrated disease similarity and the integrated miRNA similarity to construct the feature vector of the miRNA-disease pair.After feature selection,the Biased Support Vector Machine(Biased-SVM)algorithm is adopted to predict potential miRNA-disease associations.In the part of performance evaluation,we still implement local LOOCV,global LOOCV and 5-fold cross validation to evaluate the performance of PUMDA.We also use PUMDA to implement case studies on four common diseases: esophageal cancer,prostate cancer,lung cancer and lymphoma.Among the top 50 predicted disease-related miRNAs,46,43,48 and 49 miRNAs are verified by databases,respectively.In summary,it can be seen that the two models presented in this paper are effective and reliable.