Drug combination therapy is pivotal for complex diseases, but identifying synergistic three-drug regimens remains challenging due to both combinatorial explosion and the opacity of existing computational models. To address this, we introduce HyperSynergyX, an explainable framework that integrates synergy prediction with mechanistic explanation. Its core predictive component, a Dual-Biased Random Walk on Hypergraphs (DBRWH), models higher-order interactions among drugs on a three drug hypergraph and identifies latent combination patterns via tensor decomposition. To enhance interpretability, we couple DBRWH with a knowledge-graph-enhanced retrieval augmented generation (KG-RAG) module that retrieves mechanistically relevant subgraphs and uses them to generate biologically grounded hypotheses for predicted synergies. On breast-cancer data, DBRWH achieves AUROC/AUPRC of 0.9593/0.9453 under 5-fold cross-validation, and on lung cancer data it achieves 0.9262/0.9481, outperforming strong deep learning and hypergraph baselines. By linking predictive performance with mechanistic interpretability, HyperSynergyX provides a robust and transparent tool to accelerate multi-drug discovery and support rational regimen design in precision oncology. The code is available at: https://github.com/wangqi27/HyperSynergyX.
Wang et al. (Thu,) studied this question.