CurvEdgeNet: edge feature integration with Ricci curvature for circRNA-disease association prediction
Frontiers in Genetics · 2026-11-09 · 17
DOI: 10.3389/fgene.2026.1909079 ↗Abstract
Introduction Circular RNAs (circRNAs) are important regulatory non-coding RNAs involved in diverse pathological processes and have emerged as promising candidates for disease biomarker discovery. Systematic identification of circRNA–disease associations can help elucidate disease mechanisms and prioritize candidates for biological validation. However, existing computational approaches often rely on conventional similarity networks or node-centered feature learning, which may inadequately capture the non-Euclidean geometry and edge-level interactions of biological networks. Methods We propose CurvEdgeNet, a geometry-enhanced computational framework for predicting circRNA–disease associations. CurvEdgeNet constructs a heterogeneous circRNA–disease network by integrating multiple circRNA and disease similarity measures with known associations. A discrete Ollivier–Ricci curvature-based edge descriptor is incorporated into a curvature-enhanced graph attention mechanism to characterize local network geometry. In parallel, a bidirectional long short-term memory (BiLSTM) module captures global feature dependencies, while multi-head attention integrates complementary local geometric and global representations for association prediction. Results CurvEdgeNet was evaluated on three public benchmark datasets: CircR2Disease V2, CircRNADisease V2, and CircFunBase. Under five-fold cross-validation, the model achieved area under the receiver operating characteristic curve (AUC) values of 0.9854 ± 0.0031, 0.9900 ± 0.0020, and 0.9932 ± 0.0016 on the three datasets, respectively. These results demonstrate the effectiveness of the proposed framework in predicting circRNA–disease associations. Discussion Incorporating graph-theoretic curvature-based edge representations and edge-enhanced learning enables CurvEdgeNet to better characterize complex circRNA–disease association patterns. The framework provides a computational approach for prioritizing candidate disease-associated circRNAs and may facilitate further investigations into disease mechanisms and biomarker discovery.
EUVIMED