Discrimination of kimchi ingredients and identification of key metabolites of lactic acid bacteria using multivariate analysis

This study investigates the fermentation properties of eight key kimchi ingredients and identifies the major metabolites produced by ten lactic acid bacteria (LAB). Using LC-MS/MS-based non-targeted metabolomics, multivariate analysis was performed to find significant metabolites in ingredient groups. The identified key metabolites were further validated using machine learning models to confirm their reliability and relevance. The graphical abstract was designed using BioRender ( https://www.biorender.com ). • Metabolites contribute to kimchi flavor, influenced by ingredients and bacteria. • Eight vegetables and ten LAB combination formed two groups based on metabolite profiles. • Bacterial presence impacts metabolite profiles more than specific species or genera. • OPLS-DA and machine learning identified amino acids as key predictive metabolites. Kimchi is a representative, fermented vegetable food, and the bioconversion of vegetable ingredients by dominant lactic acid bacteria (LAB) is the core fermentation process of kimchi. However, naturally fermented kimchi is influenced by various bacteria and raw ingredients, making it difficult to maintain consistent quality, which limits mass production. Standardization of kimchi can address this challenge. Therefore, study of metabolic profile changes during fermentation by LABs can optimize the process by revealing correlations between bacteria and ingredients. This study may contribute to sustainable food production by improving fermentation efficiency, food security and responsible production. This study investigated metabolite changes in the major vegetable ingredients of kimchi, including kimchi cabbage, carrot, green onion, radish, onion, garlic, ginger, and red pepper, after fermentation with LAB from the Lactobacillus , Leuconostoc , and Weissella genera. Metabolites were quantified using liquid chromatography–mass spectrometry, followed by comprehensive statistical analysis. Clustering and principal component analyses showed that, despite the presence of diverse bacterial species, metabolite profiles did not exhibit clear separation by bacterial groups. Instead, samples were classified into two groups based on vegetables: group 1 (baechu, carrot, green onion, radish) and group 2 (onion, garlic, ginger, red pepper). Multivariate analyses, including PLS-DA and OPLS-DA, identified key metabolites, mainly nucleosides and amino acid derivatives, influencing kimchi’s taste. Machine-learning models validated these findings, confirming the importance of vegetable composition in determining kimchi’s sensory and functional properties.