Absolute quantification and interpretable machine learning reveal the effects of pile turning on the microecology of stacking fermentation in sauce-flavor Baijiu

Pile turning is a key step in the stacking fermentation of sauce-flavor Baijiu, yet its effects on microecological succession remain unclear. Here, we addressed this issue by integrating absolute quantification with interpretable machine learning (ML). After pile turning, the temperature and ethanol content of fermented grains increased significantly, whereas acidity decreased. The absolute abundance of Acetobacter rose rapidly, reaching 4.12 × 10 5 copies/g at 144 h, and replaced Virgibacillus as the dominant bacterial group, whereas the total absolute abundance of fungi declined. Community assembly analysis showed that bacterial communities shifted gradually from deterministic to stochastic dominance, whereas fungal communities remained dominated by stochastic processes. Moreover, machine-learning models constructed using absolute abundance data effectively distinguished samples by pile-turning status. The random forest (RF) and gradient boosting (GB) models predicted this status from bacterial and fungal communities, respectively, each achieving an AUC of 0.93. SHAP analysis further identified key discriminatory genera, including Acetobacter and Xeromyces . PICRUSt2 prediction indicated that the abundance of genes encoding enzymes involved in the production of ethanol and other alcohols was higher after pile turning, whereas the predicted abundance of functions associated with organic acid production decreased. This study provides a theoretical basis for the targeted regulation of sauce-flavor Baijiu fermentation.