Risk-Aware Computational Prioritization and Validation Route Design for Medicine-Food Homology Plant Compounds in a...

Network pharmacology studies of medicine-food homology plants have identified broad injury response pathways and hubs that cannot support compound-level or Parkinson's disease (PD)-specific claims. We developed a traceable, non-weighted framework that separates regulatory provenance, PD-context evidence, structural support, and safety/developability liabilities. A ten-plant feasibility panel was locked before overlap with a 1631-gene PD union, yielding 382 plant-associated targets and 190 strict intersections. Leave-one-plant-out analysis retained 173-190 targets, whereas disease source and threshold stress tests showed curation dependence. Whole-blood classifiers showed modest five-fold discrimination (area under the curve, 0.606-0.682) and were excluded from candidate decisions. Redocking-validated AutoDock Vina and protein-ligand interaction fingerprints retained baicalein-MMP9, baicalein-AKT1, and baicalein-BCL2 as caution-tagged follow-up pairs. Quercetin-MMP9 was retained as a liability-tagged comparator, while KCNH2 relations were safety-only. Because no biological validation is presented, these pairs remain hypotheses for prospective MPP⁺-treated SH-SY5Y testing with orthogonal injury, dopaminergic phenotypes, target dependency, material confirmation and safety controls. Baicalein remains source-pending for the material chain.