Geographic discrimination of Sauce-flavor rounded-base Baijiu using GC-MS/MS combined with multivariate analysis and machine learning.

A high-throughput targeted quantification method based on GC-MS/MS was established to determine 156 trace volatile aroma compounds (TVACs) in Sauce-flavor Baijiu. The method showed good linearity over 2-10,000 μg/L (R2 > 0.99), low LOQs (< 25.76 μg/L), satisfactory precision (RSD < 3.96%), acceptable recoveries (82.11-118.04%), and low matrix effects (≤ ±15%). Compared with HS-SPME-GC-MS, it allowed high-throughput, more accurate quantification with simple pretreatment. Based on orthogonal partial least squares discriminant analysis (OPLS-DA) combined with VIP > 1 and p < 0.05, 40 potential differential compounds were screened from 46 SRBB samples originating from six regions. Random forest model was subsequently established and evaluated using stratified 5-fold cross-validation to preserve class distribution, followed by validation with an independent test set, achieving an accuracy of 90.00%. SHAP (Shapley additive explanations) analysis identified compounds contributing to differences among regions, providing a useful approach for regional discrimination and flavor profiling of Baijiu.