Rare microbial taxa as potential drivers of yield variation in sauce-flavor baijiu fermentation: Insights from microecology and machine learning.

The multi-round, solid-state fermentation process of sauce-flavor baijiu exhibits substantial fluctuations in baijiu yield. However, the microecological factors underlying yield variation and their regulatory strategies remain unclear. Here, we systematically examined microecological structure and physicochemical parameters from rounds 1 to 7 to explain yield variation. Rare microbial taxa, particularly always rare taxa (ART) and conditionally rare taxa (CRT), played key roles in shaping microbial networks and community assembly. Moreover, rare microbes such as Lentibacillus and Pseudomonas exhibited distinct niche partitioning from abundant microbes like Acetobacter and Zygosaccharomyces (SF: 65.83%, PF: 61.89%), thereby synergistically contributing to baijiu yield variation. Furthermore, a stacking ensemble model was developed and integrated into an interactive web platform for yield prediction, achieving higher accuracy and generalization than single algorithms (R2 = 0.80-0.86; error = 6.34%). SHAP (SHapley Additive exPlanations) interpretability analysis further clarified that high-yield rounds primarily benefited from low lactic acid (< 5.10 mg/g) and low acidity (< 3.33 mg/g) in jiupei, whereas low-yield rounds were constrained by insufficient starch (< 13.23%) and moisture (< 43.82%). This study reveals the critical ecological role of rare microbes in yield variation and proposes an artificial intelligence model for fermentation control.