Effects of yeast strains and skin maceration methods on volatile aroma compounds in head, heart, and tail fractions of spine grape (Vitis davidii Foëx) wine spirit: A machine learning approach.

By multivariate analysis and explainable machine learning, this study systematically investigated the effects of two fermentation factors (yeast strain and skin maceration) on volatiles distribution in the head, heart, and tail fractions of spine grape (Vitis davidii Foëx) wine spirit. Distillation cut primarily determined volatiles distribution, whereas fermentation conditions further modulated volatile composition. Skin maceration enhanced higher alcohols, terpenes, and sulfur compounds, whereas clear-juice fermentation favored medium-chain fatty acids and their esters. The three yeast strains also differentially shaped volatile profiles, particularly regarding ester formation and terpene/norisoprenoid expression. The XGBoost-SHAP model successfully identified key discriminant volatile markers, including ethyl acetate, ethyl decanoate, β-damascenone, and 4-ethylguaiacol. The integration of chemometrics and explainable machine learning provides deeper insights into volatile marker discovery and the regulation of spine grape wine spirit composition, establishing a robust scientific basis for fermentation management and distillation cut optimization.