Screening and Identification of Ester-Producing Strains From Fengxiang Taibai Baijiu Fermentation: Machine Learning-Assisted Evaluation Under Ethanol-Acetic Acid Co-Stress.

Fengxiang Taibai baijiu (FTB) fermentation involves complex microbial interactions that critically influence aroma formation; however, the efficient selection of high ester-producing yeasts and the prediction of their fermentation performance remain challenging. In this study, yeasts were systematically isolated from different stages of FTB fermentation and evaluated using a combination of fermentation experiments and machine learning approaches. A total of 156 strains were obtained, among which seven Wickerhamomyces anomalus strains exhibited high ester-producing capacity (> 1.15 g/L). Among them, strain F12D-38 demonstrated the strongest tolerance and aroma-producing potential under ethanol-acetic acid co-stress conditions. An XGBoost model achieved high predictive accuracy (R2 = 0.89) and was applied to guide strain screening and process optimization, identifying an optimal inoculation range of 14%-16% (v/v). Fermentation validation showed that five strains outperformed the traditional control, with F12D-38 achieving an approximately 15.0% increase in total volatile compounds. Overall, this study demonstrates that integrating machine learning with experimental validation enables efficient strain selection and fermentation optimization, providing a practical strategy for improving aroma quality and process control in FTB production.