Predicting The MiRNA-disease Association Based On RNA Sequence And Heterogenous Information Sources

MicroRNAs(miRNAs)are a class of endogenous small non-coding RNA molecules whose primary function is to inhibit gene expression at the posttranscriptional level.As research has progressed,the regulation of miRNAs in many important biological processes has been identified,including human complex diseases.Recent studies have shown that many complex diseases are often accompanied by aberrant expression of miRNAs,which allows miRNAs to serve as potential biomarkers for the diagnosis of a class of diseases as well as targets for molecularly targeted therapies,thus contributing to the diagnosis,treatment and prognosis of diseases.Largescale prediction of potential miRNA-disease associations is of great significance for the detection of pathogenic mechanisms.However,considering the inherent timeconsuming and expensive of traditional in vitro experiments,more and more attention has been paid to the development of efficient and feasible computational methods to predict the potential associations between miRNA and disease.This paper focuses on RNA sequence feature extraction and multi-source data fusion.Three computational models for miRNA-disease association prediction are proposed:1.The decomposition-based miRNA-disease association prediction method MLMDA.MLMDA uses k-mer sparse matrix to extract miRNA sequence information,and combine it with miRNA functional similarity,disease semantic similarity and Gaussian interaction profile kernel similarity information.Then,more representative features are extracted from them through deep auto-encoder neural network(AE).Finally,the random forest classifier is used to effectively predict potential miRNAdisease associations.2.The computational approach based on incremental Learning and Chaos game Representation called MISSIM.The MISSIM introduces chaos game representation to extract the deep features of miRNA sequences and quantify the similarity between miRNAs.Then,utilizing incremental learning to avoid sensitivity to hyperparameter tuning and "catastrophic forgetting" when adding new data.3.The bio-network embedding based miRNA-disease association prediction method iMDA-BN.The iMDA-BN performs a network embedding representation of miRNAs and diseases from the perspective of global biological networks,and combines its attribute information to construct a predictive computing model.The iMDA-BN has three significant advantages: 1)It uses a new method to describe disease and miRNA characteristics which analyzes node representation information for disease and miRNA from the perspective of biological networks.2)It can predict unproven associations even if miRNAs and diseases do not appear in the biological network.3)Accurate description of miRNA characteristics from biological properties based on highthroughput sequence information.These three algorithms have achieved superior prediction performance in different performance evaluations.In the five-fold cross-validation evaluation,the area under the ROC curve(AUC)of the MLMDA,the MISSIM,and the iMDA-BN were 0.9172,0.9400,and 0.9145,respectively,which were higher than the average of the most advanced algorithms.In addition,the case study also proves the robustness of these three algorithms.