Disease-lncRNA Association Prediction Based On Machine Learning And Convolutional Neural Network

Long non-coding RNA(Long non-coding RNA,abbreviated as lncRNA),participates in almost all biological processes of the organism,and accounts for a large proportion of the organism's RNA.IncRNA has a strong regulatory effect on gene expression and is closely related to some human diseases.Predicting the relationship between disease and IncRNA helps to clarify the mechanism of disease,and brings some new methods of disease prevention and treatment.In the existing research on the prediction of association between disease and lncRNA,people use machine learning and network-based methods.However,the accuracy of these methods is not very high.And these methods only extract shallow features of diseases and lncRNA,it is difficult to learn deep representation features.In this paper,two disease-lncRNA association prediction methods are proposed,which are the association prediction method based on machine learning and the association prediction method based on convolutional neural networks.The first method is a disease-lncRNA association prediction method based on machine learning.The method first constructs the features of the disease through the MeSH description of the disease and the known disease-lncRNA interaction,and builds the characteristics of the lncRNA based on the assumption that the related lncRNA has similarly related diseases,thereby constructing the characteristics of the disease-lncRNA pair;then automatic encoder is used to reduce the dimensionality of the disease-lncRNA pair;finally,the forest is used to predict the disease-lncRNA association.Multiple evaluation criteria and case studies show that this method has a good effect.The second method is based on the disease-lncRNA association prediction method with convolutional neural network.The method first constructs the features of disease and lncRNA.The features of disease are divided into three parts:interaction with lncRNA,interaction with miRNA,and similarity to disease.The characteristics of lncRNA are also divided into three parts:similarity to lncRNA,interaction with miRNA,interaction with disease.Then we take two modules of convolutional neural network and machine learning,train these two modules separately.We extract the deep representation features of disease-lncRNA pairs through the convolutional neural network module,and enhance the performance of the model Through the machine learning module to further.The experimental results of cross-validation show that this method has the best effect.