Information-enhanced heterogeneity-aware model for circRNA-disease association prediction

Abstract

Circular RNAs (circRNAs) are increasingly recognized for their roles in complex diseases; however, identifying circRNA-disease associations remains challenging due to the high cost of experimental validation and the sparsity of known associations. Existing models often suffer from insufficient information and fail to effectively model heterogeneity across different data representations, including the complementary semantic and structural characteristics of multi-view data, which limits their performance. To address these issues, we propose IHCDA, an information-enhanced and heterogeneity-aware learning model for circRNA-disease association prediction. Unlike existing models that loosely combine multiple techniques, IHCDA introduces auxiliary relational structures to enhance sparse similarity information by incorporating structural relational knowledge, and seamlessly integrates it into the representation learning process. Furthermore, heterogeneous multi-view representations are adaptively modeled via view-specific attention mechanisms, and a semi-supervised contrastive learning strategy is designed to jointly optimize intra-view and cross-view relationships on the enhanced representations. Extensive experimental results demonstrate that IHCDA achieves competitive performance in circRNA-disease association prediction. Furthermore, case studies validate its effectiveness, reliability, and stability in identifying potential associations.