DualEnc-CDA: Predicting CircRNA-Disease Associations via Complementary Structural Encoding

CircRNAs have been shown to play important roles in the development of var-ious human diseases, making the identification of circRNA-disease associations an essential task for understanding disease mechanisms and discovering potential biomarkers. However, experimentally validated associations remain limited, leading to sparse and incomplete interaction data. In this study, we propose DualEnc-CDA, a novel computational framework that integrates bio-logical attribute modeling with two complementary structural encoding strategies for circRNA-disease association prediction. Specifically, shallow structural modeling captures local interaction patterns through node co-occurrence, while deep structural modeling learns higher-order dependency structures in the association network. These structural representations are fused with circRNA sequence features and disease functional similarities to form comprehensive node embeddings. Extensive experiments on three benchmark datasets demonstrate that DualEnc-CDA achieves superior and stable performance compared with existing methods. Ablation studies confirm the complementary contributions of shallow and deep structural modeling, while case studies further support the biological relevance of the predicted associations. These results indicate that DualEnc-CDA provides an effective and reliable solution for circRNA-disease association prediction.