Knowledge Graph in CircRNA-Disease Association Prediction: A Review

Abstract

Circular RNAs (circRNAs) play critical roles in biological processes and disease pathogenesis. Knowledge Graphs (KG) excel at integrating heterogeneous biomolecular data and have become powerful tools for circRNA-disease association prediction. This review aims to systematically summarize KG-based models, analyze key performance factors, existing challenges, and future directions. We classified current KG methods into knowledge embedding and graph neural networks, summarized mainstream omics databases for KG construction, and compared model performance under unified evaluation metrics and benchmarking strategies. Model performance is mainly determined by node feature completeness, network heterogeneity adaptability, and classifier selection. Inconsistent benchmark standards cause comparative bias. Major limitations include sparse validated associations, single-dimensional node features, and insufficient global learning capability. Existing KG models lack unified datasets and biological semantic fusion. Most suffer from poor interpretability and limited clinical translational potential, restricted by database sparsity and static network design. Future work should prioritize standardized dynamic KG construction, multi-modal feature integration, hybrid network design, and cross-domain fusion with large language models and transfer learning to advance circRNA-disease prediction.