Research On Pathogenic MiRNA Prediction Algorithm Based On Tensor Decomposition

As a kind of regulatory factors in human body,miRNAs function by targeting mRNAs.Dysfunction of miRNAs has an important relationship with diseases.It is significant to study the association between miRNA and diseases for the prevention and treatment of diseases.The pathogenic miRNA prediction algorithm aims to find potential pathogenic miRNAs based on known miRNA and disease-related data by computational methods.With the development of miRNA-disease prediction algorithm research and the accumulation of biological data,integrating gene information in the prediction of pathogenic miRNAs is not only helpful to improve the accuracy of prediction,but also helpful to further explore the pathogenesis of diseases.Therefore,in this thesis,two pathogenic miRNA prediction algorithms based on information related miRNA-disease association,miRNA-gene association and gene-disease information are proposed.In the first place,the existing prediction algorithms for predicting pathogenic miRNA with genetic information fusion usually apply miRNA-gene association information and disease-gene association information into two basically independent parts,which ignores the regulation mechanism of miRNA.Therefore,a tensor decomposition based pathogenic miRNA prediction algorithm,TDMDA(A tensor decomposition method for miRNA-disease association prediction)is proposed in this thesis.Firstly,the information related miRNA-disease associations,miRNA-gene associations and gene-disease associations are used to build miRNA-gene-disease association tensor.Then,tensor decomposition technique is applied to capture potential associations among miRNAs,genes,and diseases with miRNA-gene-disease association tensor and auxiliary information.Finally,convert miRNA-gene-disease association prediction results into pathogenic miRNA prediction results.The experimental results show that the performance of the TDMDA is better than other comparison methods in the prediction of pathogenic miRNAs.At the same time,the TDMDA can realize the prediction of miRNA-gene-disease associations and provide richer information for the understanding of the complex biological mechanism between miRNAs and diseases.The associations between miRNA and disease are complex and nonlinear.The miRNA-gene-disease association tensor constructed based on the associations among miRNAs,genes and diseases has a more complex structure.To better capture the nonlinear structure in the miRNA-gene-disease association tensor and identify potential pathogenic miRNAs,this paper proposes a pathogenic miRNA prediction algorithm NTDMDA(A tensor neural decomposition method for miRNA-disease association prediction)based on neural tensor decomposition.Firstly,for getting the weighted miRNA-gene-disease association tensor,the ideal of weighted K-nearest neighbor(WKNN)algorithm is utilized to update the miRNA-disease association matrix,miRNA-gene association matrix,and gene-disease association matrix with biological similarity information.Then a neural tensor decomposition algorithm is used to learn the potential non-linear structure in the miRNA-gene-disease association tensor to complete the prediction task of pathogenic miRNAs.By comparing with other advanced prediction algorithms,the prediction performance of NTDMDA is verified,and it can provide more ideas for the research of pathogenic miRNA predictive algorithms.