Predicting Non-coding RNAs And Diseases Association Based On Machine Learning

MicroRNA(miRNA)and long non-coding RNA(lncRNA)are two kinds of non-coding RNAs with different transcript lengths.Studies have found that miRNAs and lncRNAs play important roles in many human biological processes,and their disorders may lead to a variety of diseases such as cancer.Exploring the potential association between miRNAs or lncRNAs and diseases is conducive to understanding the pathogenesis of diseases and make diagnosis,treatment and prognosis timely.However,traditional biological experiments are costly and time-consuming,so it is of great significance to develop effective computational models.Machine learning methods have many applications in solving prediction problems.Done in this paper was inferring miRNA-disease association by a matrix completion model and predicting lncRNAs-disease association based on a network algorithm.For the prediction of miRNA-disease association,lncRNA-disease association data was used as the auxiliary information,and a low-rank matrix completion method was applied to make the prediction.The known miRNA-disease association data and lncRNA-disease association data were download and fused,then mapping network was built,negative samples were constructed according to the invariant properties of the mapping network,finally,the prediction of mi RNA-disease association was transformed into a low-rank matrix completion problem with similarity as the edge information,and the model was solved by an alternating gradient descent method.The result of 5-fold cross validation(5-fold CV)showed that the area under the ROC curve(the value of AUC)was 0.8884,which was exceeding some recent methods.To predict lncRNA-disease association,a random walk algorithm with restart on multiplex and heterogeneous networks was developed.First,the downloaded known lncRNA-disease association data was used to calculate various similarities to form a variety of similarity networks,next,a multiplex and heterogeneous networks were constructed with the lncRNA-disease association network.Then a random walk with restart was performed on the multiplex and heterogeneous networks to predict the potential lncRNA-disease association by using the final stable probability.The results of LOOCV showed that the value of AUC was 0.8581,which was significantly improved compared with the classical algorithm for predicting lncRNA-disease association in recent years.Finally,the models were summarized and future research were discussed in this paper.