Inferring LncRNA-disease Associations Based On Network Model And Multi-source Data Integration

Long non-coding RNAs(lnc RNAs)are a class of non-coding RNA with transcripts more than 200 nucleotides,which have the function of gene regulatory.More and more study has found that lnc RNAs involved in a various of life process,and play important role in diseases,especially the occurrence and development of cancer.Detecting the role of lnc RNAs in diseases is an important direction of current research,and the potential associations between lnc RNA and disease provide candidate biomarkers and drug targets for biological experiments.Therefore,it is a great significance to predict the associations between lnc RNAs and diseases by computational biology method.In view of the high computational complexity and much parameters of the existing methods,a novel method for lnc RNA-disease association prediction based on network model is proposed.The lnc RNA-disease heterogeneous network was constructed by combining the lnc RNA-disease association data,the disease similarity network and the lnc RNA similarity network,and then applying link prediction based on the local information to predict potential lnc RNA-disease associations.The experimental results show that the AUC of the common neighbor index is 0.849 in the leave one out cross validation,which gets best performance in the ten similarity indexes,and it is also better than the two methods based on propagation algorithm and random walk.Our method that has a good scalability can also be applied in the weighted lnc RNA-disease heterogeneous network.In addition,the lnc RNA-disease heterogeneous network and prediction result can be greatly extended after integrating the lnc RNA expression data,and seven prediction results in top ten that can be verifyed by literature mining.Therefore,the method for lnc RNA-disease association prediction based on link prediction has very good scalability,applicability and high accuracy in addition to the low computational complexity.In order to make the prediction effect not limited by the known long non-coding RNA-disease association data,we also propose another lnc RNA-disease association prediction method based the multi-source data integration.The p-value between lnc RNAs and diseases were calculated via the hypergeometric distribution after integrating four different type of data,where significant p-value means they are likely to have the potential association.The experimental results show that the integrated method achieves the best results in terms of prediction number,recall rate,F1 value and the accuracy of the top-ranked results compared with using mi RNA or gene alone.By analyzing the six cancers,most of the cancer related lnc RNAs ranked top ten have been verifyed by literature mining,which also shows our method has a strong predictive power.In this paper,two methods are proposed to predict lnc RNA-diseases associations,and satisfactory results have been obtained in both validation indexes and case studies,which are great significance in the study of diseases and lncRNAs.