A Novel Network Embedding-based Study For MiRNA-disease Association Prediction

In recent years,an increasing number of studies have indicated that there are close relationships between miRNA with disease.In particularly,miRNAs have been new biomarkers for human cancer,which is important to cancer preventions and treatments.Therefore,identifying the miRNA-disease associations has gradually become a hot topic in biology.Early traditional biological experiments identified the disease-related miRNAs by detecting the expression level of miRNAs in biological disease process.Traditional experiments achieve high accuracy,while it has the limitations of long experimental time,high cost,and low success rate.To resolve these issues,for effectively and accurately predict potential miRNA-disease associations,increasing researchers adopted computational model to predict miRNA-disease associations.This thesis proposes three computational models for miRNA-disease association prediction based on multi-source data by network embedding:1.An effective diffusion-based computational model for predicting miRNAdisease association(DF-MDA).This model firstly uses the 6)-mer algorithm to numerically represent miRNA attribute information and calculates disease semantic similarity as disease attribute information.Then,the miRNA-disease binary network is constructed based on the known miRNA-disease associations and the behavior information of each node are extracted by diffusion model.The diffusion method is easier to obtain the global characteristic information of the network than the random walk method.Finally,we constructed feature descriptors by integrating the attribute information and behavior information.These feature descriptors are trained and classified by the Random Forest classifier.2.Network embedding-based model to predict miRNA-disease associations using heterogeneous information network(HIN-MDA).At present,most researchers mainly focus on the two subjects of miRNA and disease when constructing networks.However,the molecules in the human body are common existing and influence each other.Therefore,based on the assumption that molecules are related with each other in human physiology,we construct a molecular association network used five kinds of molecules and the relationships between them.Then the network feature information of miRNA and disease is extracted by Laplacian Eigenmaps.The random forest classifier is used to score and predict the miRNA-disease association relationship.3.A Structural deep network embedding-based model for miRNA-disease associations prediction(SDNE-MDA).Most existing miRNA sequence feature information extraction algorithms only quantify one of position information or nonlinear information.To more comprehensively extract the information contained in the miRNA sequence,we use the chaos game representation(CGR)to quantify the position information and nonlinear information in the miRNA sequence.The behavior information of the node in the molecular association network were extracted by the structural deep network embedding.These information were input to the convolutional neural network to predict the miRNA-disease associations.Three different algorithms have achieved good results in performance evaluation.In terms of five-fold cross validation based on the HMDD v3.0,the average AUC of DF-MDA,HIN-MDA and SDNE-MDA achieved 0.9293,0.9319 and 0.9447,which are better than other current computational models.In addition,the results of case study also indicate the outstanding predictive performance of these three models.