Comparison of suansun fermentation methods based on SBSE-GC-MS combined with SVM machine learning.

INTRODUCTION

This study aimed to analyze the flavor profile and microbial community structure of 54 Suansun samples, fermented using three different methods: direct fermentation, natural water-sealed fermentation, and natural fermentation. The combination of SBSE-GC-MS, electronic nose, 16S rRNA, and SVM machine learning was used for comprehensive discrimination.

METHODS

The flavor components and microbial community structure were analyzed using SBSE-GC-MS, electronic nose, and 16S rRNA sequencing. SVM machine learning was employed to classify the samples based on their characteristics.

RESULTS

A total of 114 common aroma components were identified, including esters, alcohols, hydrocarbons, ketones, acids, aldehydes, heterocyclic compounds, phenols, halogenated hydrocarbons, amides, and others. Using a p < 0.05 and VIP > 1 threshold, 27 key characteristic flavor compounds were identified, with the highest concentration found in the natural water-sealed fermentation method. The SVM model achieved a 100% discrimination rate. Dominant bacterial genera identified across the methods were Lactiplantibacillus, Lactococcus, Weissella, and Limosilactobacillus, with a 95.65% match between dominant genera and key flavor compounds in natural water-sealed fermentation.

DISCUSSION: The study highlights that natural water-sealed fermentation is the most effective method for enhancing flavor profiles, and that Weissella plays a significant role in the production of key flavor compounds, particularly p-cresol, which increased over 600 times in natural water-sealed fermentation. Direct fermentation significantly shortens the fermentation cycle, while natural water-sealed fermentation offers the best results in terms of flavor development.