Decoding the Flavor Structure of Jiang-Flavor Low-Alcohol Base Baijiu: A Machine Learning-Driven Approach to Reveal the Flavor Evolution Patterns and Key Quality Control Nodes.

Low-alcohol base baijiu, a by-product of Jiang-flavor baijiu production, contains abundant flavor compounds, yet its flavor structure remains insufficiently understood, limiting its high-value utilization. This study integrated flavoromics with interpretable machine learning to characterize the flavor structure of Jiang-flavor low-alcohol base baijiu and identify its key driving factors. During distillation, ethanol decreased from 38.67% to 6.50% (v/v), whereas total acidity increased from 2.74 to 7.53 g/L and pH decreased from 3.01 to 2.59, accompanied by a sensory transition from harmonious and full-bodied notes to sour and salty attributes. Based on sensory and flavor component analysis, the samples can be divided into three gradients according to ethanol volume fraction (>20%, 10-20%, <10%), and their flavor profiles exhibit a systematic evolution: as alcohol content decreases, the relative content of esters and alcohols decreases, while the relative content of acids increases. Random Forest identified 3-methyl-1-butanol, furfuryl ethyl ether, and tetramethyl pyrazine as key markers positively associated with harmonious and smooth attributes (r > 0.93) and negatively associated with sour and salty notes. Odor activity value analysis further quantified stage-dependent sensory contributions. The XGBoost model accurately predicted flavor stages (accuracy = 0.955), and SHAP analysis highlighted low-threshold esters and acids as critical regulators of flavor quality. These findings provide a quantitative and interpretable framework for precision blending, quality grading, and the valorization of low-alcohol base baijiu.