Prediction Of LncRNA-disease Associations Based On Heterogeneous Network Method

In recent years,with the rapid development of biological science and continuously innovative biological experimental technology,an increasing number of evidence shows that long non-coding RNA(lncRNA)plays a vital role in biology and is closely related to the occurrence of human diseases.About 12% of lncRNAs can be found in other organisms except human beings.H19 and Xist were the first two lncRNAs to be found.These two lncRNAs were found in the early 1990 s by traditional gene location methods.Then more and more lncRNAs were found in human beings and many animals and were recorded and studied.These large numbers of lncRNA can be arranged into a complex network of gene expression.Their mutations and dysregulations are related to different human diseases and abnormal development: such as breast cancer,colorectal cancer,lung cancer and cardiovascular disease,etc.lncRNA is often used as a biomarker in biomedicine to provide a biological reference for doctors' diagnosis and treatment.Therefore,predicting the potential association of lncRNA-disease has become a crucial research hotspot in the field of human complex diseases.The existing methods are mainly divided into three aspects: network-based method,matrix decomposition method and machine learning method.The network-based method has the following disadvantages:(1)few methods can predict which lncRNA was related to new diseases;(2)only consider the relationship between lncRNA and diseases.This paper proposes two new methods based on heterogeneous networks for predicting the potential association of lncRNA-disease.The first method is a bi-random walks method based on the two networks(BiRWLD).BiRWLD constructed the lncRNA similarity network and disease similarity network,by integrating the lncRNA similarity and disease similarity.Finally,the lncRNA-disease association was predicted based on the random walk method.The second method is a random walk with restart method based on three networks(MHRWR).MHRWR integrates the similarity network of lncRNA,disease and gene with the known correlation network of lncRNA-disease,lncRNA-gene and gene-disease,and designs a global multi-layer heterogeneous network;then it uses the similarity matrix of disease to pre fill the original lncRNA-disease adjacency matrix,which is used to predict the lncRNA related to new diseases.Finally,it based on random walk with restart method to potential association of lncRNA-disease.This paper evaluates the prediction performance of the model based on leave-one-out cross-validation.Experimental results show that the AUC value of BiRWLD is 0.90183,and the AUC value of MHRWR is 0.91344,which is significantly better than some previous methods.In order to further verify its performance,the article uses BiRWLD and MHRWR models to verify lncRNAs related to colon,colorectal,and lung adenocarcinomas in case studies.