Research On The Prediction Model Of Disease-related Long Non-coding RNA Based On Multi-element Network Information

In recent years,more and more experimental results show that long noncodingRNA(lncRNA)has an important influence on various life processes.Its function attenuation and variation are closely relevant to the emergence and progress of many complex human diseases.Therefore,it is of great significance to explore the diagnosis,treatment and prognosis of human diseases.Considering the time and effort of traditional biological experiment methods to identify the potential lncRNA,researchers began to construct prediction model to explore the relationship between potential lncRNA and diseases.This paper mainly explores the potential association of lncRNA through the multi-element network constructed by the information of lncRNA,miRNA and disease from different databases.Firstly,we described the background knowledge and significance of this research interest,then the research foundation of this model is introduced.Three calculation models based on biological network information for identifying potential lncRNA of disease are proposed,and the specific contents are as follows:(1)First of all,a computational model named IIRWR is constructed to predict the potential lncRNA of human diseases.IIRWR proposes the definition of disease group,designs a method to calculate the weight score of the related disease groups of lncRNA nodes,and adds the score to the walk network formed by lncRNA nodes.Then,the network restart random walk to get the latent correlation score between lncRNA node and disease node.The experimental results show that the prediction performance of IIRWR is greatly improved than other models.(2)In view of the limitations that IIRWR cannot apply to isolated nodes,a prediction model FVTLDA based on association feature information is proposed to infer disease potentially relevant lncRNAs.FVTLDA combines restart random walk algorithm and the definition of group in model IIRWR,extracts the characteristic information of each pair of diseases and lncRNA from multi-element network,which makes the model not only avoid the dependence on known lncRNA disease association information,but also can be used to predict isolated biological nodes,and solve the problems existing in model IIRWR.In addition,the first mock exam is to avoid the limitations of the single model.FVTLDA combines linear regression and neural network to analyze the feature information.The comparative experiments and case analysis results show the superior performance of the model.(3)Considering the limitations of the model FVTLDA to extract feature information manually,this paper proposes a learning framework called CNMCLDA to automatically extract hidden biometric information from multi-element networks,and then identify the potential lncRNA.CNMCLDA adopts matrix filling form,which includes two convolutional neural networks to extract lncRNA and disease hidden features respectively,and designs five different loss function synchronous optimization model parameters.In particular,the model also designs an algorithm to divide negative samples to ensure the balance of positive and negative samples,avoiding the limitations of some traditional models.The Cross-Validation results and case analysis show that CNMCLDA is significantly better than the existing classical calculation model.Finally,this paper summarizes three prediction models in this paper and makes a prospect for the future research plan.