Prediction of LncRNA-Disease Associations Based on Network Representation Learning

Massive observations have indicated that long non-coding RNAs (lncRNAs) are crucial in a number of biological processes and associated with various human diseases. Developing an efficient calculation model to predict the associations between lncRNA and diseases is not only beneficial to disease diagnosis, treatment, prognosis and potential drug targets in drug discovery, but also avoid the waste of human and material resources brought by biological experiments. In this paper, we proposed a novel prediction of lncRNA-disease associations based on complex and comprehensive molecular associations network (MAN), which integrated nine kinds of interactions among five molecules, including lncRNA, miRNA, disease, drug and protein. Network embedding Node2vec method was applied to extract behavior feature from MAN to generate a low-dimension vector containing nodes and edges information. After implementing 5-fold cross validation, the proposed method yielded good prediction performance with an average Accuracy of 91.91%, Sensitivity of 94.05%, Specificity of 89.76%, Precision of 90.21%, MCC value of 83.91%, AUC value of 0.9746 and AUPR of 0.9693. Comparative experiment indicates the behavior feature extracted by Node2vec is more representative than attribute features of lncRNA adopted 3-mer and diseases extracted by semantic similarity. Moreover, breast cancer, colon cancer and lung cancer are explored in case study. As a results, more than half of top 5 interactions are successfully confirmed for each disease by other datasets. Based on these reliable results, it is anticipated that proposed model is feasible and effective to predict lncRNA-disease associations at a global molecules level, which is a new respective for future biomedical researches.