AGDFCDA: Adaptive graph convolutional network and dual feature for circRNA-disease association prediction

Abstract

Circular RNA (circRNA) is a special type of RNA molecule whose structure presents as a closed loop. Numerous studies have demonstrated that abnormal expression of circRNA is closely associated with the development of diverse diseases. Accurately predicting the association between the circRNA and disease is important for understanding the pathogenesis of disease and discovering potential biomarkers. However, the high cost and complexity of traditional biological experiments limit the development of research. By constructing computational models and performing bioinformatics analysis, it is possible to identify disease-related circRNA more efficiently and reveal its potential mechanism. This paper presents AGDFCDA, a computational model for circRNA-disease association prediction, featuring a dual feature extraction strategy. On the one hand, the strategy applies the fully connected neural network to reduce the redundant information in the initial features, while the hidden information of circRNA and disease is preliminarily extracted. On the other hand, the strategy introduces adaptive graph convolutional network to learn more comprehensive representation of circRNA and disease to realize further extraction of features. AGDFCDA is assessed using five-fold cross-validation, and the results indicate that it outperforms the comparison methods in predicting circRNA-disease associations. In addition, the results of case studies can provide reliable candidate circRNA for wet experiments to be carried out with effective cost savings.