Disease-associated MiRNA Prediction Based On Non-negative Matrix Factorization

With the development of microarray and next-generation sequencing techniques,a great amount of data has been generated.As one of the most important factors in gene expression regulation,microRNA has become the focus and hot spot of the research nowadays.As a matter of fact,evidences have shown that micro RN As are related with the pathogenesis of all kinds of complex diseases.Therefore,the analysis of microRNA is of great significance and value.In this thesis,our main focus is on the prediction of disease-associated miRNAs,and we propose two methods to analysis the disease-miRNA associations based on non-negative matrix factorization and its improved version with sparse constraints,which provides a new way for predicting disease-related micro RN As.The main work of this paper is as follows:(1)We propose a new method NMF to predict the disease-associated miRN As based on non-negative matrix factorization.It fully utilizes the principle of non-negative matrix factorization to integrate multiple sources of data to the prediction of miRNA-disease associations,including miRNA-gene data,disease-gene data,miRNA-miRNA functional similarity data,phenotype-disease data and known miRNA-disease data.By assuming a common basis matrix existing between miRNA-gene and disease-gene data,we can establish the relationship between miRNAs and diseases.Moreover,by using the miRN A functional similarity data,phenotype-disease data and known other mRNA-disease data as constraints during the iteration process to guide the proposed algorithm to converge as fast as possible.Experiment results show that our method has superior performance over existing methods on many types of disease datasets.(2)We propose an improved algorithm based on NMF by adding sparse constraints on non-negative matrix factorization to predict miRNA-disease associations.By imposing sparse constraints on common basis matrix and factor matrices,we further reduce the search space of the proposed algorithm.We compared the improved algorithm with NMF on the same datasets used in the third chapter.Experiment results show that,comparing with NMF,non-negative matrix factorization with sparse constraints can gain better performance and higher prediction accuracies on the same disease type.