Research On MiRNA-disease Association Prediction Based On Graph Convolution Network

mi RNAs affect disease by negatively regulating gene expression,so there is a potential association between mi RNAs and disease.Computational models are an important way to predict mi RNA-disease associations,which can reduce the cost of mi RNA-disease association analysis.Existing methods only focus on predicting binary associations and are difficult to capture high-order topological features.How to improve the prediction effect and further predict the association type is crucial for subsequent mi RNA and disease prediction research.The thesis takes mi RNA and disease association as the research object,and conducts model research from three aspects: similarity network reconstruction,mi RNA and disease association prediction,mi RNA and disease association type prediction,and designs and implements a mi RNA and disease association prediction platform.The main work of the paper is as follows:1.Aiming at the problem of sparse similarity network and noise in the process of mi RNA-disease association prediction,a similarity network reconstruction model based on symmetric three-factor non-negative matrix decomposition was proposed.Regularity constraints,solved using additive gradient descent.The simulation results show that the model has a normal mean absolute error of 0.0016 and a root mean square error of 0.1425 on the mi RNA similarity dataset,and has a normal mean absolute error of 0.0052 and a root mean square error of 0.1160 on the disease similarity dataset.The area under the receiver operating curve of the within-and between-group scoring models for mi RNAdisease association prediction based on reconstructed similarity networks was 0.8362.2.Aiming at the problem that high-order topological features of mi RNA-disease association network are difficult to capture,a mi RNA-disease association prediction model based on two-dimensional hypergraph convolutional network is proposed.Based on similarity matrix and known mi RNA-disease association network,a hypergraph is constructed.Convolutional neural networks migrate to hypergraphs to predict mi RNAdisease associations through joint factor solving.The simulation results show that the area under the receiver operating curve of this model is 0.9256 and 0.9211 in global leaveone-out cross-validation and five-fold cross-validation,respectively.3.Aiming at the problems that the network topology features of mi RNA and disease association types are difficult to obtain and the similarity information is not fully utilized,a prediction model of mi RNA and disease association types based on decomposed heterogeneous graph neural network is proposed.Based on similarity matrix and known mi RNA and disease association types The network constructs a heterogeneous graph,introduces similarity edge weights,adaptive combining functions and type features for message aggregation and node update,and uses Dis Mult for link prediction.The simulation results show that the top1 F1 on the four types of data sets are 0.6684,0.6619,0.6618,and 0.6536,respectively,and the areas under the receiver operating curve are0.9202,0.8735,0.9090,and 0.9116,respectively.4.Using browser/server architecture,based on Vue front-end,Spring Boot framework and My SQL database,the mi RNA-disease association prediction platform is designed and implemented to meet the application scenarios of mi RNA-disease association prediction.The platform includes functions such as user management,data analysis,and correlation prediction.