Research On Disease Related-MiRNA Prediction Based On Similarity Calculation And Convolutional Neural Network

MicroRNA(miRNA)is a small,endogenous,single-stranded and noncoding RNAs,which is composed of 20-24 nucleotides.Currently,many studies have shown that miRNAs have been linked to numerous human diseases closely.In addition,there is a strong relationship between miRNAs and phenotypes.On this basis,considering the possibility of using miRNA as a biomarker to mark different human diseases.And,using the association between miRNAs and human disease to help researchers gain insight into the molecular mechanisms of disease etiology.And then make an important contribution to the development of new drugs and the preventive treatment of related diseases.Therefore,using bioinformatics to explore the relationship between each miRNA and each human disease has become a focus issue.According to the calculation method based on similarity calculations,the prediction model can be divided into two categories: one is based on the known association relationship to build the network,which relies on high-quality biological networks;the other is based on machine learning lacks negative sample set to trainning.This paper will integrate multi-source biological data based on similarity calculation method to build high-quality similarity data,and combine with convolution neural network to explore the unknown miRNA disease relationship,and proposed two calculation model.First,the prediction method of the relationship between miRNA and disease based on positive sample learning.The hidden features extracted from positive samples come from the mapping of miRNA and disease in low dimensional space.The mapped miRNA and disease potential feature vectors are used to construct the feature expressions for each miRNA-disease relational pair.Using convolutional neural networks to complete the learning and prediction of relational pairs,finally.Secondly,considering the key role of gene in the miRNA disease interaction,the three-layer network of disease gene miRNA is obtained by adding the interaction network of gene disease relationship,target gene miRNA and gene-gene association network into the construction of miRNA-disease network.The feature expressions of each miRNAdisease association pair is obtained by mining the topological structure information and potential information between nodes.In 5-fold cross validation test,The experimental results show that the AUC values of the two model are achieved 90.34% and 90.82%,respectively,which are higher than other advanced algorithms.In addition,use two common human diseases: lung cancer and heart failure as case study.The top-50 prediction results of model can be supported by relevant literature.At the same time,the two model can not only be used to predict the relationship between miRNA and disease,but also can be used to the relationship between miRNA and phenotype predicition task.