Coupling ensemble learning with multi-omics: a novel data-driven strategy for Daqu quality assessment and validation.

Daqu, a fermentation starter for traditional fermented foods, is produced by complex microbial communities; it is generally evaluated using conventional methods based on subjective sensory perception and limited physicochemical indicators. These approaches have several limitations, including poor consistency and low repeatability, making it challenging to meet industrial demands for a transition from "experience-driven" to "science-driven" quality control. In this study, medium-high temperature Daqu for strong-aroma-type Baijiu was selected as the research object, and we aimed to establish a quality evaluation model for it by the integration of multidimensional fermentation data. Additionally, multi-omics analysis was combined to validate key functional microorganisms, thereby revealing their mechanisms of action. The results indicate that the fermentation functions of Daqu are predominantly driven by eukaryotic microorganisms characterized by "low abundance and high expression." Rhizomucor and Saccharomycopsis were identified as core contributors to hydrolases. An ensemble learning model (temperature-based Gradient Boosting + physicochemical XGBoost + flavor-based Linear Regression) was established to achieve rapid quality grading of Daqu. Daqu quality was substantially improved by targeted inoculation with core functional fungi, thereby verifying the effectiveness of the model. Accordingly, in this study, a systematic research paradigm of "multi-omics analysis → machine learning modeling → microbial targeted validation" was established, allowing the transformation of experience-dependent "ecological art" into interpretable, quantifiable, and predictable "synthetic ecological engineering." The study findings are expected to provide a generalizable methodology for addressing quality evaluation issues in traditional fermented foods.