Predicting LncRNA-miRNA Interactions via Network Embedding with Integrated Structure and Attribute Information

Accumulating evidence demonstrated that microRNAs (miRNAs) and long noncoding RNAs (lncRNAs) are related with some complex human diseases. LncRNA-miRNA interactions (LMIs) play an important role in regulatory of gene networks. However, the biological experiments for detecting lncRNA-miRNA interactions are often expensive and time-consuming. Thus, it is urgent to develop computational method for predicting LMIs. In this paper, we propose a novel computational approach LMMAN to predict potential lncRNA-miRNA associations based on molecular associations network (MAN). More specifically, the known relationships among miRNA, lncRNA, protein, drug and disease are firstly integrated to construct a molecular association network. Then, a network embedding model LINE is employed to extract network behavior features of lncRNA and miRNA nodes. Finally, the random forest classifier is used to predict the potential lncRNA-miRNA interactions. In order to evaluate the performance of the proposed LMMAN approach, five-fold cross-validation tests are implemented on benchmark dataset lncRNASNP2. The proposed LMMAN approach can achieve the high AUC of 0.9644, which is obviously better than the existing methods. The promising results reveal that LMMAN can effectively predict new lncRNA-miRNA interactions and can be a good complement to relevant biomedical fields in the future.