Climatic factors and initial microbiota-driven prediction of the accumulated microbiota at the end of stacking fermentation for soy sauce aroma type baijiu.

Most traditional fermented foods are produced through spontaneous fermentation, which contain complex microbiota. However, it is still unclear what the key driving factors of the microbiota are and how to predict the microbiota at the end of fermentation. This study employed a typical spontaneous fermentation process, stacking fermentation for soy sauce aroma type baijiu, as a case to predict the microbiota at the end of fermentation based on key driving factors. The microbiota at the start of fermentation was mainly provided from the starter (daqu), air, and the ground of the fermentation room, contributing 45.22%, 10.05%, and 7.48% of the bacteria, and 41.83%, 10.77% and 1.21% of the fungi, respectively. Average air temperature and average photosynthetic photon flux density were identified as key climatic factors driving microbial succession. Finally, a Transformer-based machine learning model was constructed to predict the microbiota at the end of fermentation based on key climatic factors and the initial microbiota. 29 out of 34 dominant genera (average relative abundance ≥1%) could be accurately predicted (NRMSE <30%). Aitchison distance between true and predicted value using Transformer model was significantly lower than that using Random Forest model (P < 0.001). This work revealed the roles of climatic factors and the microbiota at the start of fermentation in shaping the microbiota at the end of fermentation. The constructed predictive model could facilitate the regulation of the microbiota in food fermentation.