Predicting circRNA–Disease Associations through Multisource Domain-Aware Embeddings and Feature Projection Networks

Recent studies have highlighted the significant role of circular RNAs (circRNAs) in various diseases. Accurately predicting circRNA–disease associations is crucial for understanding their biological functions and disease mechanisms. This work introduces the MNDCDA method, designed to address the challenges posed by the limited number of known circRNA–disease associations and the high cost of biological experiments. MNDCDA integrates multiple biological data sources with neighborhood-aware embedding models and deep feature projection networks to predict potential pathways linking circRNAs to diseases. Initially, comprehensive biometric data are used to construct four similarity networks, forming a diverse circRNA–disease interaction framework. Next, a neighborhood-aware embedding model captures structural information about circRNAs and diseases, while deep feature projection networks learn high-order feature interactions and nonlinear connections. Finally, a bilinear decoder identifies novel associations between circRNAs and diseases. The MNDCDA model achieved an AUC of 0.9070 on a constructed benchmark dataset. In case studies, 25 out of 30 predicted circRNA–disease pairs were validated through wet lab experiments and published literature. These extensive experimental results demonstrate that MNDCDA is a robust computational tool for predicting circRNA–disease associations, providing valuable insights while helping to reduce research costs.