Publication

Abstract : The advancements in the field of high throughput analysis show abnormal expression of long non-coding RNAs (lncRNAs) in many complex diseases. Accurately identifying the disease association of lncRNA is essential in understanding their role in disease mechanism and subsequent therapy. The contemporary methods for predicting lncRNA-disease association use heterogeneous information learned from different biological sources such as lncRNAs, miRNAs, and diseases. However, learning topological features from diverse network structured data is one of the limiting factors of these methods. To address this challenge, we propose a method for lncRNA-disease association prediction based on Deep Belief Network (DBN), referred to as DBNLDA. In this method, three interaction networks such as lncRNA-miRNA similarity (LMS), disease-miRNA similarity (DMS), and lncRNA-disease association (LDA) network are constructed. A new framework based on the node embedding, DBN, and a neural network regression model is used to learn network and local representation of lncRNA-disease pairs. From the node embedding matrices of LMS, DMS, and LDA networks, lncRNA-disease features are learned by DBN layers. These DBN features are used to predict the association score by an ANN regression model. Compared to several state-of-the-art methods, DBNLDA obtained better AUC (0.96) and AUPR (0.967) under five-fold cross-validation. Case studies on breast, lung, and stomach cancer also affirmed the ability of DBNLDA in predicting potential lncRNAs associated with various diseases.