Targeted urinary metabolomics combined with machine learning to identify biomarkers related to central carbon metabolism for IBD.

Introduction: Inflammatory bowel disease (IBD), comprising Crohn's disease (CD) and ulcerative colitis (UC), is a chronic and relapsing inflammatory disorder of the gastrointestinal tract. Current diagnostic approaches are invasive, costly, and time-consuming, underscoring the need for non-invasive, accurate diagnostic methods. Methods: We conducted a targeted metabolomic analysis of 49 metabolites related to central carbon metabolism in urinary samples from individuals with IBD and control group. Diagnostic models were constructed using six machine learning algorithms, and their performance was evaluated by cross-validated area under the receiver operating characteristic curve (AUC). The SHAP (SHapley Additive exPlanations) method was used to interpret the models and identify key discriminatory features. Results: Six metabolites-xylose, isocitric acid, fructose, L-fucose, N-acetyl-D-glucosamine (GlcNAc), and glycolic acid-differentiated UC from control group, while three metabolites-xylose, L-fucose, and citric acid-distinguished CD from control group. The optimal diagnostic model achieved a mean AUC of 0.84 for UC and 0.93 for CD. These models retained high diagnostic accuracy even after adjusting for disease activity. SHAP analysis identified L-fucose, xylose, and GlcNAc as important features for UC, and citric acid and xylose for CD. Discussion: Our findings highlight distinct metabolic signatures in central carbon metabolism associated with IBD subtypes. The identified metabolite panels, combined with machine learning models, offer promising non-invasive tools for differentiating UC and CD from healthy individuals.