Machine learning-enhanced flavoromics: Identifying key aroma compounds and predicting sensory quality in sauce-flavor baijiu.
Journal:
Food chemistry
PMID:
39952173
Abstract
The quality of Sauce-flavor baijiu hinges on sensory characteristics and key aroma compounds, which traditional methods struggle to evaluate accurately and effectively. This study explores the sensory characteristics and aroma compounds of Sauce-flavor baijiu across different rounds using flavoromics and machine learning, constructing quality grade prediction models. Sensory characteristics shift from acid in the early stages BJ1-BJ2 rounds to sauce in the mid-stages BJ3-BJ5 rounds and caramel in the late stages BJ6-BJ7 rounds. Employing AEDA and OAV analyses, 18 key odor-active compounds were identified, such as ethyl butyrate, ehyl isovalerate, and phenethyl acetate. Additionally machine learning models combined with clustering algorithms achieved high accuracy in predicting quality grades: 85 % (MLP+ HCA), 97 % (XGBoost+ K-means), and 84 % (Random Forest+ GMM). The SHAP model identified 20 key aroma compounds, including diethyl succinate, Tetramethylpyrazine, and Acetaldehyde, determining quality concentration thresholds. This study offers robust methods for baijiu flavor control and quality evaluation.