Identification of age-specific urinary metabolic biomarkers in Wilson disease using machine learning: a comparative...

The diagnosis of Wilson disease (WD) is complicated by heterogeneous clinical phenotypes and inadequate performance of routine biomarkers. This study sought to screen age-specific urinary metabolic biomarkers to assist WD diagnosis via ensemble tree-based machine learning algorithms.

Sixty participants were enrolled, comprising 30 WD patients (10 pediatric, 20 adult) and 30 healthy controls. Morning urine samples were analyzed using UPLC-Q-TOF-MS in both ionization modes. Orthogonal partial least squares discriminant analysis identified differential metabolites. Four ensemble algorithms (Random Forest, GBDT, XGBoost, LightGBM) were compared using nested cross-validation to minimize overfitting risk. LASSO regression selected optimal metabolite panels. Model performance was evaluated using 5-fold cross-validation with AUC as the primary metric.

XGBoost achieved AUC values of 0.87±0.03 (pediatric) and 0.96±0.02 (adult) in nested validation, with permutation testing confirming performance above chance levels (p>0.001). Sixty-eight differential metabolites were identified in pediatric patients versus 109 in adults, with 15 metabolites consistently altered across both age groups. Pathway analysis revealed age-specific disruptions: nucleotide metabolism in pediatric patients and oxidative stress pathways in adults. LASSO feature selection identified 7 metabolites for pediatric and 3 for adult classification while maintaining high diagnostic accuracy.

This study successfully developed the first age-specific metabolomics-based diagnostic approach for WD using ensemble machine learning. The distinct metabolic signatures between age groups provide mechanistic insights into WD pathophysiology and support personalized diagnostic strategies for improved patient outcomes. These exploratory findings require independent validation before clinical implementation.