KGAACDA: Knowledge graph attention-aware algorithm for CircRNA-disease association prediction.

Recent research highlights the critical involvement of circRNAs (circular RNAs) in the pathogenesis of various human diseases. Consequently, the efficient identification of circRNA-disease associations (CDA) holds significant potential for advancing disease diagnostics and therapeutic strategies. However, traditional experimental validation methods are often costly and time-consuming, leading to the development of computational approaches as promising alternatives. To address the persistent challenges of data sparsity and the effective capture of intricate underlying associations from complex circRNA-disease interaction networks, this paper introduces a ‌Knowledge Graph Attention-Aware Algorithm for CircRNA-Disease Association (KGAACDA)‌ prediction. The algorithm first constructs initial feature embeddings for nodes based on multi-dimensional similarity metrics for circRNAs and diseases. It also introduces an optimized negative sampling strategy informed by these similarity measures to enhance model training. Subsequently, a comprehensive ‌knowledge graph (KG)‌ is constructed by integrating diverse biological databases, thereby leveraging rich heterogeneous information to mitigate data sparsity issues. The KG is then processed using a multi-head attention mechanism that dynamically weights different types of biological relationships, thereby providing greater biological interpretability for the associations learned between heterogeneous nodes. Higher-order features extracted from the KG and the initial node embeddings are then fused through a Multi-Layer Perceptron (MLP), with the model's parameters being optimized using a composite loss function. Our extensive experimental evaluations demonstrate that ‌KGAACDA outperforms current state-of-the-art baseline algorithms‌ in terms of AUC (Area Under the Receiver Operating Characteristic Curve) and AUPR (Area Under the Precision-Recall Curve) on two independent datasets, establishing it as a robust and effective computational tool for predicting CDA.