Machine-Learning-Assisted Aroma Profile Prediction in Five Different Quality Grades of Nongxiangxing Baijiu Fermented During Summer Using Sensory Evaluation Combined with GC×GC-TOF-MS.

Flavor is one of the crucial factors that influences the quality and consumer acceptance of baijiu. In this study, we analyzed the volatile organic compound (VOC) profiles of five different quality grades of Nongxiangxing baijiu (NXB), fermented during the summer of 2024, using 2D gas chromatography time-of-flight mass spectrometry (GC×GC-TOF-MS). We employed machine-learning (ML)-based classification and prediction models to evaluate the flavor. For TW, the scores of the sensory evaluation for coordination and overall evaluation were the highest. TW contained the highest concentration of ethyl caproate; we detected 965 VOCs in total, including several pyrazine compounds with potential health benefits. Principal component analysis (PCA) combined with orthogonal partial least squares discriminant analysis (OPLS-DA) enabled us to distinguish different samples, with eight VOCs emerging as primary contributors to the aroma of the samples, possessing variable importance in projection (VIP) values > 1. Furthermore, we tested eight ML models; random forest (RF) demonstrated the best classification performance, effectively discriminating samples based on their VOC profiles. The key VOC contributors that showed quality-grade specificity included 1-butanol, 3-methyl-1-butanol, and ethyl 5-methylhexanoate. The results elucidate the flavor-based features of NXB and provide a valuable reference for discriminating and predicting baijiu flavors.