Prediction Of LncRNA-disease Association Based On Heterogeneous Network And Machine Learning

lnc RNAs(long noncoding RNAs)are kind of noncoding RNAs with more than 200 nt nucleotides in length.There are increasing evidences showing that lnc RNAs play key roles in many biological processes,and mutions in and dysregulation of them have been proven to be correlated with a broad range of human diseases.However,experimental methods to identify associations between lnc RNAs and diseases are expensive and timeconsuming.Effective computational approaches can improve our understanding of the molecular mechanisms of human diseases and aids in finding biomarkers for disease diagnosis,treatment,and prevention.This paper focused on computational methods to predict lnc RNA-disease associations.Biological network-based methods have been applied to many fields of bioinformatics.In this paper,the lnc RNA-lnc RNA similarity matrix and disease-disease similarity matrix were obtained using the known lnc RNA-disease association,lnc RNA expression similarity and disease semantic similarity.The lnc RNA-disease heterogeneous network was constructed by similarity matrices,and the bi-random walk algorithm was performed on the herogeneous network to predict unknown associations.The leave-one-out cross validation and 5-fold cross validation were implemented to evaluate the predicted performance,it obtained reliable AUCs of 0.9374 and 0.8504 respectively,showing the best performance compared to several computation methods.Also,heterogeneous network-based models have shown good results in predicting highassociated diseases and predicting new diseases.Futher case studies were implemented on the menthod,illustrating the effectiveness of the model.Besides,the method can be applied to a disease with no known associated lnc RNA.Machine learning methods have many applications for solving problems such as recommendation systems.Using the known lnc RNA-disease associations,by introducing mi RNA and disease-related information,a new lnc RNA similarity matrix was calculated and the data set was extended through the relationship between diseases.By introducing stacked autoencoders and multilayer neural networks,a neural network-based machine learning model which can capture high-level features of data and an ensemble learning model makes use of multiple classifiers are proposed.5-fold cross validation were implemented,and two models achieved an AUC of 0.9041 and 0.9072 on the new dataset,respectively.The model can also predict mi RNA-disease associations and achieve good results.After adding some noise data to the dataset,the model can still achieve good results,illustrated model can be robust when noise data were introduced.Case studies of prostate,lung and grastic cancer futher demonstrate the ability of model to predict unknown lnc RNA-disease associations.In addition,the model can predict for diseases without any known associations.This paper proposed heterogeneous network-based method and machine learning-based methods to predict lnc RNA-disease associations.Good results are obtained in both experimentally relevant indicators and case studies.Therefore,the method may be a good complement to future biomedical research.In the last,the paper summarized proposed models and prospected future research.