Answers are grounded in this paper and cite their sources.
Accurate identification of drug–target interactions (DTIs) is fundamental to drug discovery and repositioning. In this study, DTI prediction is formulated as a binary classification task, in which each drug–protein pair is classified as interacting or noninteracting. Existing heterogeneous graph‐based approaches often rely on predefined interaction edges, which can limit information propagation and hinder the modeling of complex molecular relationships, particularly for isolated drugs and targets. To address these challenges, we propose KARA‐DTI, a dual‐encoder framework incorporating an adaptive routing mechanism to enhance information propagation in drug–target heterogeneous graphs. By introducing global router nodes, KARA‐DTI facilitates information exchange between otherwise disconnected entities and alleviates the isolated‐node problem. In addition, a Kolmogorov–Arnold Network (KAN)‐based graph encoder captures nonlinear molecular representations, while a bidirectional cross‐attention module models fine‐grained interactions between drug and protein features. Experiments on four benchmark datasets, namely KIBA, C. elegans, BindingDB, and Human, demonstrate that KARA‐DTI achieves competitive performance across multiple evaluation metrics, including AUROC, accuracy, AUPRC, sensitivity, and F1‐score. In particular, KARA‐DTI achieves the best results on multiple metrics across the evaluated datasets, demonstrating the effectiveness of its adaptive routing and cross‐modal feature fusion mechanisms for drug–target interaction prediction.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: