Prediction Of MiRNA-disease Association Based On Restart Random Walk And Stacked Autoencoder

MicroRNA(miRNA)is a non-codingRNA composed of about 18-25 nucleotides.It participates in cell life processes such as cell metabolism,proliferation,apoptosis and development,it is also involved in the occurrence,development and treatment of many human diseases,and becomes a key regulatory factor in cell function.However,the number of miRNAdisease associations that have been discovered is very small,which limits the role of miRNA in the field of disease treatment.Therefore,studying miRNA-disease associations has important theoretical value and application significance in the field of human disease research and treatment.The calculation methods of miRNA-disease associations prediction mainly include network methods,machine learning methods and deep learning methods.The network methods use a variety of biomolecular information to construct miRNA and disease similarity networks,which enrich the feature information that affects miRNA-disease associations prediction.Machine learning methods use feature selection algorithms to select features,which improves the prediction performance of miRNA-disease associations.Deep learning methods can extract the deep level features of data through layer-by-layer abstraction,but the existing deep learning methods only consider miRNA and disease similarity networks,and do not consider that the selection of the optimizer affects the prediction performance of the model to a certain extent.In view of the existing problems in the field of miRNA-disease associations prediction,this paper constructs a miRNA-disease associations prediction method based on restart random walk and Stacked Autoencoder.The innovations and main research contents of this paper are as follows:(1)The existing methods fail to consider the topological structure information of miRNA and disease similarity networks when predicting miRNA-disease associations.In order to solve this problem,this paper proposes a miRNA-disease associations prediction method based on topological structure of network.Firstly,restart random walk method and positive pointwise mutual information method are used to obtain the topological structure information of each miRNA and disease network.Secondly,the topological similarity matrixes are input into the Stacked Autoencoder to obtain the dense low dimensional features of miRNAs and diseases.Finally,the extracted features are input into the deep neural network to predict the associations between miRNAs and diseases.The experimental results show that capturing the topological structure information of miRNA and disease similarity networks is helpful to predict miRNAdisease associations.(2)In order to solve the problem that model SAEMDA has the problem of inaccurate optimization direction when optimizing the deep neural network,a miRNA-disease association prediction method named R-SAEMDA based on Radam optimization is proposed in this paper.R-SAEMDA introduces Radam optimizer into model to optimize the loss function in the process of deep neural network training.Radam optimizer adopts variance correction method to reduce the variance of parameter update in the early stage of model training,and alleviate the problem of inaccurate optimization direction of deep neural network at the initial stage of training.The experimental results show that Radam optimizer performs better than other optimizers,and the overall prediction performance of the model is improved after adopting Radam optimizer.To sum up,this paper considers the topological structure information of miRNA and disease similarity networks,uses Stacked Autoencoders to extract low-dimensional features of miRNAs and diseases,introduces the Radam optimizer to optimize the training process of deep neural network,and predicts miRNA-disease associations.Compared with existing research,the R-SAEMDA model has better predictive performance and provides auxiliary information for exploring the role of miRNA in the field of disease treatment.