Machine Learning-Based Profiling of Urinary Volatile Organic Compounds Reveals Distinct Diagnostic Signatures for Gastric Cancer.

INTRODUCTION: Gastric cancer (GC) is a leading cause of cancer-related mortality worldwide, often diagnosed at advanced stages due to the absence of specific symptoms. This study aimed to explore the potential of urinary volatile organic compounds (VOCs) as non-invasive biomarkers for gastric cancer diagnosis. METHODS: Urine samples from 60 GC patients and 65 healthy controls (HC) were analyzed via gas chromatography-ion mobility spectrometry (GC-IMS), detecting 84 VOCs. Machine learning algorithms-LASSO, SVM-RFE, and Boruta-were applied to select diagnostic biomarkers. Predictive models were constructed using the H2O AutoML framework, with the dataset randomly partitioned into training and test sets at an 7:3 ratio. The training set underwent 5-fold cross-validation to develop and optimize the models, and the resulting optimal model was validated on the reserved test set using metrics including the area under the curve (AUC), sensitivity, and specificity. RESULTS: A total of 84 VOC peaks were detected in the urine samples from GC patients. The OPLS-DA model successfully distinguished between the GC and control groups, indicating distinct metabolic features. The LASSO model demonstrated superior diagnostic performance, identifying a panel of six characteristic VOCs with high sensitivity and specificity. Moreover, a stacked ensemble model achieved an AUC of 0.93 (95% CI: 0.87-0.99) in the training set and 0.92 (95% CI: 0.83-1.00) in the test set, with 100% sensitivity (95% CI: 82.4%-100.0%) and 84% specificity (95% CI: 64.0%-94.8%), demonstrating substantial value for GC diagnosis. CONCLUSIONS: Our findings suggest that urinary VOCs may serve as promising non-invasive biomarkers for the diagnosis of gastric cancer. The identification of these VOCs provides valuable insights into the metabolic alterations associated with GC and highlights the potential for integrating VOC profiling into clinical practice for improved diagnostic strategies. Further validation in larger cohorts is warranted to confirm the clinical applicability of these findings.