Prediction Of MicroRNA-disease Association Based On Matrix Factorization And Deep Learning

In recent years,a large number of studies have shown that the study of microRNA is helpful to the prevention,diagnosis and treatment of complex human diseases.Micro RNA participates in many key biological processes,and microRNA is also the pathogenesis of complex human diseases,which indicates that identifying the association between microRNA and diseases can help us understand diseases at microRNA level.Therefore,revealing the potential relationship between microRNA and diseases is an important topic in biomedical field.Checking all possible microRNA-disease associations by traditional biological experiments is time-consuming and laborious,and predicting potential microRNA-disease associations can provide important prior information for medical researchers,therefore,researchers pay more and more attention to developing an feasible and effective calculation method for predicting the potential association between diseases and microRNA.First of all,we use many different methods to calculate the similarity between microRNA and diseases in this study,which including semantic similarity,sequence similarity,functional similarity,Gaussian kernel similarity and Hamming distance similarity,etc.In order to integrate the four similarities,we propose a new iterative fusion method to reduce the sparsity of the original microRNA-disease association matrix,we preprocess the original association matrix.Secondly,the influence of similarity matrix fusion,data preprocessing and parameter adjustment on several matrix factorization methods is compared and analyzed.It provides reference for subsequent model calculation steps,parameter setting and other researchers.A new matrix factorization model is constructed.Hessian regularization,cooperative regularization and L_2,1norm are added to the matrix factorization model,and the model is solved by gradient descent method,the AUC of this model is 0.9235 under5-fold cross validation.Finally,a hybrid model framework based on deep learning,matrix factorization and singular value factorization is proposed.Extracting features from both linear and nonlinear perspectives.The ACC of this model is 0.967 under the 5-fold cross validation.In addition,we have made relevant case analysis on three important human diseases:lung tumor,bladder tumor and breast tumor.Of the Top-50 microRNA predicted by the model,48,45 and 46 have been confirmed in relevant databases and literatures respectively,which indicating that the prediction ability of the matrix factorization model is reliable.