HPLC-HRMS and interpretable machine learning decipher serum lipidomic signatures in NSCLC.

Background: Non-small cell lung cancer (NSCLC) remains the leading cause of cancer mortality, largely due to the lack of reliable non-invasive tools for detection and risk stratification. Lipid metabolic reprogramming is a hallmark of cancer and may serve as a promising source of diagnostic biomarkers. Methods: Serum from 40 NSCLC patients and 30 controls was profiled by high-performance liquid chromatography-high-resolution mass spectrometry (HPLC-HRMS), quantifying 331 annotated lipids. Differential and pathway analyses were performed. Least absolute shrinkage and selection operator (LASSO), support vector machine (SVM), Extreme Gradient Boosting (XGBoost), and Light Gradient-Boosting Machine (LightGBM) models were evaluated using stratified 10-fold cross-validation; feature prioritization used recursive feature elimination and Shapley additive explanations (SHAP). A combined clinical-lipid model incorporating selected lipids and clinical covariates was assessed with discrimination, calibration, and decision-curve analysis. Results: NSCLC exhibited broad decreases in glycerophospholipids, sphingolipids, and triacylglycerols, consistent with membrane-lipid remodeling. LightGBM showed the best discrimination in internal validation. Key discriminant lipids included lysophosphatidylcholine (LPC(O-18:1)), decanoylcarnitine, and sulfatide (SL) (SL 38:5). The integrated lipid-clinical model achieved good discrimination (area under the receiver operating characteristic curve (AUC) = 0.946) and acceptable calibration. A nomogram was constructed for individualized risk estimation. Conclusions: This study nominates candidate serum lipid markers and an interpretable modeling workflow for NSCLC classification in an exploratory case-control cohort. External validation and targeted quantification in larger, multicenter and screening-relevant populations are required before clinical implementation.