Xuan Ping, Li Haoyuan, Cui Hui, Xu Zelong, Nakaguchi Toshiya, Zhang Tiangang
IEEE J Biomed Health Inform. 2025 Aug 20;PP. doi: 10.1109/JBHI.2025.3600406.
As circular non-coding RNA (circRNA) is closely associated with various human diseases, identifying disease-related circRNAs can provide a deeper understanding of the mechanisms underlying disease pathogenesis. Advanced circRNA-disease association prediction methods mainly focus on graph learning techniques such as graph convolutional networks and graph attention networks. However, these methods do not fully encode the multi-scale neighbor topologies of each node, and the dependencies among the pairwise attributes. We propose a multi-scale neighbor topology-guided transformer with Kolmogorov-Arnold network (KAN) enhanced feature learning for circRNA and disease association prediction, termed MKCD. The model integrates multi-scale neighbor topology, complex relationships among multiple nodes, and the global and local dependencies of pairwise attributes. First, MKCD incorporates an adaptive multi-scale neighbor topology embedding construction strategy (AMNE), which generates neighbor topologies covering varying scopes of neighbors by performing random walks on a circRNA-disease-miRNA heterogeneous graph. Second, we design a dynamic multi-scale neighbor topology-guided transformer (DMTT) that leverages the multi-scale neighbor topologies to guide the learning of relationships among circRNA, miRNA, and disease nodes. The multi-scale neighbor topology is dynamically evolved, providing adaptive guidance to the transformer's learning process. Third, we establish a feature-gated network (FGN) to evaluate the importance of topological features derived from DMTT and the original node attributes. Finally, we propose an adaptive joint convolutional neural networks and KAN learning strategy (ACK) to learn the global and local dependencies of circRNA and disease node pair features. Comprehensive comparison experiments show that MKCD outperforms six state-of-the-art methods, improving AUC and AUPR by at least 14.1% and 7.6%, respectively. Ablation experiments further validate the effectiveness of AMNE, DMTT, FGN and ACK innovations. Case studies on three diseases further validate the application value of our method in discovering reliable circRNA candidates for diseases of focus. The source code and datasets are freely available at https://github.com/pingxuan-hlju/MKCD.
由于环状非编码RNA（circRNA）与多种人类疾病密切相关，识别与疾病相关的circRNA有助于更深入地了解疾病发病机制。先进的circRNA-疾病关联预测方法主要集中在图学习技术上，如图卷积网络和图注意力网络。然而，这些方法没有充分编码每个节点的多尺度邻域拓扑结构以及成对属性之间的依赖关系。我们提出了一种具有柯尔莫哥洛夫-阿诺德网络（KAN）增强特征学习的多尺度邻域拓扑引导变压器，用于circRNA和疾病关联预测，称为MKCD。该模型整合了多尺度邻域拓扑、多个节点之间的复杂关系以及成对属性的全局和局部依赖关系。首先，MKCD采用了一种自适应多尺度邻域拓扑嵌入构建策略（AMNE），通过在circRNA-疾病-miRNA异构图上进行随机游走，生成覆盖不同邻域范围的邻域拓扑。其次，我们设计了一种动态多尺度邻域拓扑引导变压器（DMTT），利用多尺度邻域拓扑来指导circRNA、miRNA和疾病节点之间关系的学习。多尺度邻域拓扑动态演化，为变压器的学习过程提供自适应指导。第三，我们建立了一个特征门控网络（FGN）来评估从DMTT导出的拓扑特征和原始节点属性的重要性。最后，我们提出了一种自适应联合卷积神经网络和KAN学习策略（ACK），以学习circRNA和疾病节点对特征的全局和局部依赖关系。综合比较实验表明，MKCD优于六种先进方法，分别将AUC和AUPR提高了至少14.1%和7.6%。消融实验进一步验证了AMNE、DMTT、FGN和ACK创新的有效性。对三种疾病的案例研究进一步验证了我们的方法在发现针对重点疾病的可靠circRNA候选物方面的应用价值。源代码和数据集可在https://github.com/pingxuan-hlju/MKCD上免费获取。
