Deciphering Spatiotemporal Dynamics of Fermented Grains in the Jiangxiangxing Baijiu Production Process: Insights From a Transformer-Based Deep Learning Model.
Aug 2026· Journal of Food Science· Vol 91 8, pp.
e71316
· 0 citations· 19 references
Medicine
TL;DR
The capability of deep learning to decode complex fermentation data is demonstrated and this framework supports intelligent process monitoring and quality control, with potential applicability to other complex food fermentation systems.
Abstract
The solid-state fermentation of Jiangxiangxing Baijiu exhibits marked spatiotemporal heterogeneity in microbial communities and physicochemical parameters. We characterized the microbial succession and physicochemical dynamics of Zaopei (fermented grains) across Da-hui rounds (Rounds 3-5) and applied LimiX, a Transformer-based deep learning framework, for fermentation state monitoring. Spatial stratification explained 99% of microbial community variance (PERMANOVA, P < 0.001), with inner layers harboring greater bacterial diversity than surface layers. Staphylococcus and Weissella correlated negatively with starch and reducing sugar, while Oceanobacillus and Acetobacter strongly discriminated among rounds. LimiX achieved an AUC of 1.000 for round classification and outperformed GLM and Random Forest (RF) in discriminating fermentation layers. External validation yielded 100% accuracy for round differentiation and 92.86% for temporal stage prediction. SHAP analysis identified Klebsiella and Thermoascus as the primary drivers of spatial and temporal stratification, respectively. These results demonstrate the capability of deep learning to decode complex fermentation data and provide a theoretical and technical basis for the intelligent digitalization of Baijiu production. PRACTICAL APPLICATIONS: A Transformer-based deep learning model was developed to integrate amplicon sequencing data and physicochemical indices for decoding the spatiotemporal dynamics of fermented grains in Jiangxiangxing Baijiu fermentation. This framework supports intelligent process monitoring and quality control, with potential applicability to other complex food fermentation systems.
Sauce-flavor Baijiu is a Chinese distilled spirit with a highly nonlinear, multivariable, and coupled pit fermentation process, complicating real-time parameter acquisition and regulation. In this study, a deep neural network (DNN) based multi-task framework was developed using industrial-scale microbial community profiles and physicochemical parameters, enabling prediction of fermentation states and in silico optimization of controllable parameters to regulate target flavor compounds. Firstly, microbial communities, physicochemical parameters, and volatile flavor compounds were analyzed across seven production rounds. The results suggested that dominant microorganisms included Acetilactobacillus, Kroppenstedtia, Pichia, Monascus, and Kazachstania. Across rounds, acidity increased while pH decreased. Total esters and alcohols increased from rounds 1-3, then declined toward round 7, with ethyl acetate and phenethyl alcohol being key discriminants. Using this seven-round fermentation-parameter dataset, a DNN framework was developed with two functions: (i) fermentation parameter prediction and (ii) targeted flavor regulation via an optimization module coupled to the trained model. Baseline comparisons and ablation experiments demonstrated that the model achieved good predictive performance (Pearson's r = 0.801; R2 = 0.690; CCC = 0.783). Evaluation using data from a subsequent production year showed encouraging predictive agreement for key parameters. As a concept validation, the DNN-guided ethyl acetate regulation strategy was evaluated in a simulated round-6 pit fermentation by inoculating with Companilactobacillus pabuli, Fructilactobacillus fructivorans, and Pichia kudriavzevii. This intervention increased ethyl acetate and ethanol levels, supporting the feasibility of the model-guided intervention direction. Overall, this work supports intelligent monitoring and flavor-oriented control in sauce-flavor Baijiu production, providing a foundation for automated manufacturing.
Daqu is the fermentation starter of traditional Chinese vinegar production. However, its quality is location-specific, experience-dependent, and batch-differential. This study aims to unveil the physiochemical and microbial dynamics of Daqu during fermentation and improve the qualities of Daqu as well as vinegar through engineering the microbiota. Layered and mixed Daqu samples were collected at six fermentation stages and subjected to physiochemical and microbiota analysis. Correlation analysis revealed that the Daqu microbiota-driving factors changed from daily average temperature and contents of water and soluble protein to accumulated temperature, pH, and reducing sugar content, resulting in the higher activities of acidic protease, amylase, cellulase, and xylanase at early stages and higher glucoamylase, pectinase, and esterase activities at late stages. Null-model inference based on the beta-nearest taxon index (βNTI) and the Bray-Curtis-based Raup-Crick metric (RCbray) revealed different assembly patterns in Daqu communities, with stronger deterministic assembly of bacterial communities and ecological drift in fungal communities. The core microbiota consisted of molds (Aspergillus, Thermoascus, Rhizopus, Rhizomucor, and Rasamsonia), yeast (Saccharomycopsis), actinomycete (Saccharopolyspora), and bacilli (Bacillus, Lactilactobacillus, Leuconostoc, and Weissella). Using the culturomics strategy, 443 distinct strains confined into 40 genera and 74 species were isolated, and 59 of them showed significant protease and/or carbohydrase activities. A synthetic microbial consoria (SynCom) of five strains (Aspergillus oryzea, Saccharomycopsis fibuligera, Lichtheimia ramosa, Bacillus velezensis, and Lactilactobacillus plantarum) was prepared referring to the core microbiota composition and hydrolytic performance, which supplementation significantly improved the Daqu quality as well as the fermentation efficacy for vinegar brewing, including higher activities of glucoamylase and protease, higher yields of alcohol, organic acids, and amino acid nitrogen, and better sensory properties. The SynCom-modified microbiome with low microecological vulnerability and upregulated metabolism of carbohydrates, amino acids, and lipids might be the underlying reason. This study advances the Daqu microbiome research and offers valuable ideas to engineer the Daqu microbiota for improved qualities of Daqu and vinegar.
Rui Ma, Xiaojie Nan, Yanting Li et al.· Food Research International· 0 citations
Mechanized production of Maotai-flavor Baijiu (MFB) is increasingly adopted in the Baijiu industry; however, microbial succession and flavor-related metabolic potential throughout the complete eight-round mechanized stacking fermentation (SF) process remain insufficiently understood. In this study, microbial communities, functional genes, physicochemical properties, and volatile compounds during SF were investigated using metagenomic sequencing and headspace solid-phase microextraction coupled with gas chromatography-mass spectrometry (HS-SPME-GC/MS). A total of 168 volatile compounds were detected, of which 41 representative compounds were selected for further analysis. Among them, 15 differential volatiles were identified by PLS-DA, with furfural showing the highest abundance. Microbial profiling revealed pronounced community differentiation and continuous succession across fermentation rounds. Acidity, starch, and reducing sugars were significantly associated with microbial community variation, with acidity and starch exhibiting the strongest associations. In the initial round (R1), microbial communities were mainly derived from raw materials and Daqu. Bacterial communities shifted from lactic-acid-bacteria-enriched communities to those characterized by Kroppenstedtia and Bacillus, whereas fungal communities transitioned from yeast-enriched stages to mold-enriched and mold-yeast coexistence stages. Metagenome-inferred functional annotation, co-occurrence network, and correlation analyses suggested potential links between microbial succession and flavor-related metabolic pathways. Yeasts were mainly associated with ethanol- and organic-acid-related metabolism during the early stage, whereas Bacillus and Kroppenstedtia were linked to predicted starch-degradation and organic-acid-related pathways during the middle and late stages. Overall, this study provides a comprehensive characterization of microbial succession and metagenome-inferred flavor-related metabolic potential during mechanized SF and offers reference data for process monitoring and quality management in MFB production.
Shuyi Wan, Wenrui Huang, Zongjie Zhang et al.· Journal of food microbiology· 0 citations
The microbial fermentation of mulberry branch residues offers a potential strategy for sustainable lignocellulosic biomass valorization. This study integrated 16S rRNA gene sequencing and untargeted LC–MS metabolomics to characterize temporal microbial and metabolic changes during a 12-day solid-state fermentation at 35 °C using a defined consortium of lactic acid bacteria, Bacillus subtilis, and Saccharomyces cerevisiae. Microbial community analysis revealed distinct temporal succession, with Bacillus accounting for 19.5% of the bacterial community during early fermentation, followed by increasingly diverse assemblages. Exploratory machine learning analysis ranked Azotobacter and Kyrpidia among the genera contributing most strongly to temporal differentiation. FAPROTAX-based predictions indicated that chemoheterotrophy-related functions remained prevalent across fermentation stages, although these predictions do not represent direct functional activity. Untargeted metabolomics detected 2782 putatively annotated features, dominated by lipids (15.1%), organic acids (13.8%), and phenylpropanoids (13.2%). Correlation analysis identified temporal associations between bacterial genera and metabolite classes, including associations of Geobacillus with alkaloids and glycerophospholipids. Exploratory OPLS-DA further highlighted pseudouridine and 3-methylxanthine as discriminatory features across fermentation stages. These findings provide a descriptive multi-omics overview of microbial succession and metabolic variation during mulberry branch residue fermentation.
The spontaneous fermentation of Zhejiang rosy vinegar (ZRV) is driven by environmental microbiota, but the processes underlying its flavor formation remain poorly understood. Using metagenomic sequencing, we investigated microbial community assembly, environmental drivers, and metabolic networks during industrial-scale ZRV fermentation. Acetic acid dominated the final organic acids. Community assembly shifted toward deterministic selection with rising acidity, with a slight rebound of stochastic processes in the late stage (R2 values of 0.442 and 0.346 for bacteria and fungi, respectively). Mantel tests confirmed that environmental factors significantly regulated microbial assembly. Co-occurrence networks grew more complex, with positive interactions accounting for 85.24% (bacteria) and 90.10% (fungi) in the late stage. Key genes (ldh, gapA, pgk) from Acetobacter pasteurianus and Lactobacillus acetotolerans dominated late-stage fermentation, while genes (adhP, SDH) from Aspergillus oryzae and Saccharomyces cerevisiae supported early- and mid-stage fermentation. These findings elucidate microbiota-driven metabolic pathways in ZRV, supporting the fermentation window optimization and industrial vinegar quality standardization.
Cheng-Hao Jin, Shi-rui Liu, Wenlan Song et al.· Journal of Agricultural and...· 0 citations
Medium-temperature Daqu (MTD) exhibits pronounced spatial heterogeneity between its surface (QS) and inner (QI) portions, yet their functional differences remain poorly understood. This study compared physicochemical properties, metabolites, and species-level microbial communities of QS and QI and evaluated their impacts on simulated strong-flavor Baijiu (SFB) fermentation. QS exhibited higher saccharification and liquefaction activities and higher counts of lactic acid bacteria and yeasts, whereas QI was enriched in thermophilic molds and volatile compounds. Thirty-two microbial biomarkers differentiated QS and QI, and QS showed a more complex microbial co-occurrence pattern. Simulated fermentation revealed early bacterial divergence followed by convergence to Acetilactobacillus jinshanensis. During mid-to-late fermentation, QI showed higher levels of acids and esters than QS. Collectively, QS was mainly associated with substrate conversion and early fermentation initiation, whereas QI was more closely associated with flavor development during later fermentation stages. These findings provide valuable insights into MTD optimization and SFB quality control.
