Discovery of a preliminary urinary metabolite panel for Parkinson’s disease: a pilot study using paired patient-spouse samples and machine learning consensus

Introduction Parkinson’s disease (PD) lacks reliable non-invasive diagnostic biomarkers. Urine is a promising biofluid for biomarker discovery, but the profound influence of shared environment and lifestyle represents a major confounder. Methods To rigorously address this, we designed a pilot study using a unique matched-pair cohort: PD patients together with their healthy spouses. Untargeted LC–MS metabolomics was performed on urine samples from 15 carefully matched pairs. Differential features were identified using VIP > 1.0 and p < 0.05. A multi-model consensus strategy (RF, SVM, PLS-DA) was applied to prioritize robust candidates from 2,640 annotated metabolites, followed by filtering for pharmacological relevance and collinearity. Results A preliminary five-metabolite panel (Cyanuric acid, Benzeneacetonitrile, 3-Formylsalicylic Acid, dADP, and ent-cassa-12,15-dien-2beta-ol) was defined. Despite the inherently small sample size of this niche cohort, the panel demonstrated promising internal discriminative performance (AUC > 0.95). Discussion We emphasize that these results are exploratory. The primary contribution of this pilot study is not a validated diagnostic tool, but the demonstration of a carefully controlled design to isolate PD-specific metabolic signatures and the proposal of specific candidate biomarkers. This work establishes a critical proof-of-concept and prioritizes targets for essential future validation in larger, independent cohorts.