AMPGLDA: Predicting LncRNA-Disease Associations Based on Adaptive Meta-Path Generation and Multi-Layer Perceptron

Abstract

Increased evidence suggests that long non-coding RNA (lncRNA) holds a vital position in intricate human diseases. Nonetheless, the current pool of identified lncRNA linked to diseases remains restricted. Hence, the scientific community emphasizes the need for a reliable and cost-effective computational approach to predict the probable correlations between lncRNA and diseases. It would facilitate the exploration of the underlying mechanisms of lncRNA in ailments and the development of novel disease treatments. In this study, we propose a novel approach for predicting the associations between lncRNAs and diseases, which relies on the adaptive meta-path generation (AMPGLDA). Firstly, we integrate information about lncRNA, diseases, and miRNAs to construct a heterogeneous graph. Then, we utilize principal component analysis to extract global features from nodes. Based on this heterogeneous graph, AMPGLDA adaptively generates multiple meta-path graph structures and uses a graph convolutional neural network to learn the semantic feature representations of lncRNA and disease from the meta-path. Ultimately, AMPGLDA utilizes a deep neural network classifier to accurately predict the association between lncRNA and disease. The AMPGLDA model achieves impressive results, with AUC and AUPR scores of 99.66% and 99.66%, respectively, under the independent test set. Furthermore, three case studies demonstrate its accuracy in discovering new lncRNA-disease associations.