Stage-dependent lipid remodeling during traditional Zhenba bacon processing: Insights from untargeted lipidomics and explainable machine learning.

Traditional bacon processing involves marked lipid remodeling, yet the stage-dependent dynamics of individual lipid species remain unclear. In this study, fatty acid analysis, UPLC-MS/MS-based untargeted lipidomics, and interpretable machine learning were integrated to characterize lipid transitions during Zhenba bacon processing. Polyunsaturated fatty acids, especially linoleic acid, increased significantly during curing, suggesting pronounced lipid mobilization at this stage. Untargeted lipidomics identified 34 core differential lipids associated with processing-stage discrimination. K-means clustering revealed three temporal patterns: continuous decline, peak at curing, and progressive relative enrichment. Some structural phospholipids and triglycerides decreased progressively, whereas several complex became relatively enriched during late processing. AdaBoost-SHAP further identified TG(60:8) and PS(47:2) as representative lipid features contributing strongly to stage classification. These findings demonstrate that lipid remodeling during Zhenba bacon processing is highly stage-dependent and non-linear, and provide a lipid-based basis for process monitoring and quality evaluation.