Machine learning-driven multi-omics integration of urinary organic acids and ions enables precision risk...

Calcium oxalate (CaOx) nephrolithiasis is closely associated with metabolic dysregulation, while current risk assessment based on 24-h urine analysis is time-consuming and inconvenient. This study aimed to develop a noninvasive predictive model for CaOx stones using morning urine organic acid and inorganic ion profiles combined with machine learning, and to identify metabolomic biomarkers related to CaOx stone formation.

A total of 232 CaOx stone formers and 238 healthy controls were enrolled. Organic acids and inorganic ions in morning urine were quantified by gas chromatography-mass spectrometry and ion chromatography, respectively. Participants were randomly divided into training and testing sets (8:2). Diagnostic models were constructed using random forest, support vector machine, logistic regression, and extreme gradient boosting, with 10-fold cross-validation for optimization. Model performance was evaluated using AUC, accuracy, sensitivity, specificity, F1-score, and G-mean. Differential metabolite screening based on p-values and fold change, together with SHAP-based feature prioritization across multiple machine learning models, was integrated to identify candidate metabolites. The selected metabolites, together with BMI, were then incorporated into a multivariable logistic regression model to construct a nomogram.

No significant differences in age or sex were observed between groups, whereas BMI was higher in the CaOx group (p < 0.05). Twenty differential metabolites were identified (|log₂FC| > log₂ [1.5], p_adj < 0.05). SHAP analysis consistently highlighted seven metabolites across algorithms. Integration of differential and SHAP-based selection yielded five key metabolites, which, combined with BMI, produced a model with an AUC of 0.8439 (95% CI: 0.8071-0.8806), outperforming any single indicator.

This study integrates morning urine organic acid and inorganic ion profiling with machine learning to establish a predictive model for CaOx nephrolithiasis. Five urinary metabolites were identified as potential biomarkers, providing a convenient tool for risk assessment and new insights into CaOx stone pathogenesis.