MiRNA-related Prediction Research Based On Ensemble Learning And Personalized Recommendation

Although microRNAs(miRNAs)cannot encode protein,they can regulate gene expression and have important functions in various life activities of the human body.Studies have shown that miRNAs are closely related to the occurrence and development of many human diseases,and miRNAs are feasible targets for small molecule drugs.The study of disease-ralated and small molecule drug-related miRNAs is of great significance for not only mechanism understanding of disease occurrence and development,but also improvement of disease diagnosis and drug development.The miRNA-disease and small molecule drug-miRNA association prediction models could identify the most possible associations,which have important guiding effect in biological experiments.We constructed models to predict disease-related and small molecule drug-related miRNAs based on ensemble learning and personalized recommendation methods,respectively.The main research contents are as follows:(1)We proposed a computational model named Ensemble of Decision Tree based MiRNA-Disease Association prediction(EDTMDA).First,miRNA features and disease features are constructed from miRNA-disease association information and integrated similarity information.Then,by randomly selecting negative samples and features including miRNA features and disease features,as well as implementing feature dimensionality reduction,multiple decision trees are constructed.Finally,the average of their predicted scores is used as the miRNA-disease association score.Compared with the single learner method,ensemble learning has higher prediction accuracy.Moreover,feature dimensionality reduction is incorporated into ensemble learning to remove noise and redundant information in the learning process and also reduce computational complexity of model.(2)We proposed a computational model named Bayesian Ranking for MiRNADisease Association prediction(BRMDA).Based on the characteristics of the miRNAdisease association prediction research,BRMDA is constructed by improving the Bayesian personalized ranking(BPR)algorithm from three aspects: 1)adding miRNA similarity information and disease similarity information;2)adding miRNA bias term;3)adopting neighbor information-based method to predict associations for new diseases or new miRNAs.As a personalized recommendation method,BRMDA can generate optimal miRNA ranking results for each disease.(3)We proposed a computational model named Ensemble of Kernel Ridge Regression based Small Molecule drug-MiRNA Association prediction(EKRRSMMA).First,the integrated small molecule similarity and miRNA similarity is used as small molecule drug features and miRNA features,repectively.Then,based on the random selection of features and feature dimensionality reduction,multiple kernel ridge regression learners are constructed from the perspective of small molecule drug and miRNA,respectively.Finally,average of their predicted scores is used as the small molecule drug-miRNA association score.For the proposed miRNA-disease association and small molecule drug-miRNA association prediction models,we evaluated their performace by implementing multiple cross validation methods and case studies.The results show that the performance of these models is very reliable.