Abstract
Identifying effective drug combinations for multidrug therapies in intractable diseases remains challenging, and drug synergy depends on regulatory mechanisms. Herein, we introduce a mechanistic framework that redefines drug combination strategies into two distinct synergy paradigms-"boost" and "complement"-and present a network-based method called SYNERGIE to predict associated combinations across diseases. For diseases lacking known therapeutic targets, we introduced network controllability analysis to estimate therapeutic pathways and target molecules. SYNERGIE integrates drug-disease and drug-drug interactions across multiomics layers using Bayesian optimization to prioritize disease-state-specific combinations. SYNERGIE outperformed existing methods across 14 diseases, and identified both doublets and triplets. Therapeutic effects of triplets predicted for colorectal cancer were validated through in vitro experiments and multi-resolution omics analyses of human cohort data at bulk and single-cell levels. SYNERGIE offers a mechanism-aware approach to identifying clinically viable combination therapies, and is expected to support stratified treatment of intractable diseases.