Prediction Of MicroRNA-Disease Association Based On Graph Semi-supervised Learning

MicroRNA(miRNA) is a kind of micro RNA composed of about 20 to 24 nucleotides,which plays an important role in the regulation of human gene expression.Previous studies have shown that miRNA is closely related to the occurrence and development of a variety of complex diseases,so mining the potential relationship between diseases and miRNA can not only provide new clues for exploring the pathogenesis of complex diseases,but also guide clinical drug development in the future.In this paper,the research content is to uncover the potential association between miRNA diseases.We proposed four methods to predict miRNA-disease associations,A Novel Method for Mi RNA-Disease Association Prediction based on Global Linear Neighborhoods(GLNMDA),A novel method for miRNA-disease association prediction based on matrix completion and label propagation(MC|LPMDA),A novel semi-supervised model for miRNA-disease association prediction based on L1-norm graph(L1GMDA)and Adaptive multi-view multi-label learning for identifying disease-associated candidate miRNAs(AMVMLMDA).The four methods all use the functional similarity number network of miRNA,the semantic similarity network of disease and the association network of human miRNA and disease,and all use the semi-supervised learning method to predict the association based on the heterogeneous graph model of miRNA and disease.Of course,the difference between the four methods is very clear.In order to solve the problem of data noise,GLNMDA uses the global linear reconstruction algorithm to reconstruct the functional similarity data of miRNA and the semantic similarity data of disease,and then uses the label propagation algorithm to complete the prediction task.Considering that Gaussian kernel similarity is an effective method to measure the similarity between miRNA and disease,MCLPMDA uses matrix recovery algorithm to solve the problem of incomplete data set,then calculates the Gaussian kernel similarity of miRNA and disease respectively,then integrate the Gaussian kernel similarity and the complete similarity and the similarity before recovery,and finally uses label Propagation algorithm to predict the relations between miRNA and disease.Because L1 norm can produce sparse resolution,the objective function based on L1 norm can provide more reliable prediction results for predicting miRNA-disease associations.Therefore,we propose a semi-supervised model using L1 norm constraint.L1 GMDA we first recalculate the miRNA functional similarities as well as the disease semantic similarities based on the latest version of Me SH descriptors and HMDD,and then update the similarity matrices and association matrix iteratively in both miRNA space and disease space.The optimized association matrices from each space are combined together as the final output.In order to solve the problem of data noise in the current database,AMVMLMDA will learn a new similarity graph adaptively,and then get the prediction results according to multi-label learning.The results of cross validation and case studies show that these four methods can be used as an effective tool to mine miRNA-disease associations.