Machine Learning and Multi-Omics Integration to Reveal Biomarkers and Microbial Community Assembly Differences in Abnormal Stacking Fermentation of Sauce-Flavor Baijiu.

Abstract

Stacking fermentation is critical in sauce-flavor Baijiu production, but winter production often sees abnormal fermentations, like Waistline and Sub-Temp fermentation, affecting yield and quality. This study used three machine learning models (Logistic Regression, KNN, and Random Forest) combined with multi-omics (metagenomics and flavoromics) to develop a classification model for abnormal fermentation. SHAP analysis identified 13 Sub-Temp Fermentation and 9 Waistline microbial biomarkers, along with 9 Sub-Temp Fermentation and 12 Waistline flavor biomarkers. Komagataeibacter and Gluconacetobacter are key for normal fermentation, while Ligilactobacillus and Lactobacillus are critical in abnormal cases. Excessive acid and ester markers caused unbalanced aromas in abnormal fermentations. Additionally, ecological models reveal the bacterial community assembly in abnormal fermentations was influenced by stochastic factors, while the fungal community assembly was influenced by deterministic factors. RDA analysis shows that moisture significantly drove Sub-Temp fermentation. Differential gene analysis and KEGG pathway enrichment identify metabolic pathways for flavor markers. This study provides a theoretical basis for regulating stacking fermentation and ensuring Baijiu quality.