食品工业科技2025，Vol.46Issue(15)：36-48,13.DOI:10.13386/j.issn1002-0306.2024100221
基于机器学习研究乳酸菌发酵石榴汁的风味品质差异形成机制
Optimization of Flavor Quality of Lactic Acid Bacteria Fermented Pomegranate Juice Based on Machine Learning
摘要
Abstract
In this study,the highest weighted preferences score(HWPS)and the lowest weighted preferences score(LWPS)were selected from 90 fermented pomegranate juice(FPJ)samples according to weighted preference score.Furthermore,key volatile compounds that influenced sensory preferences were predicted in 2 FPJs by using headspace solid phase micro-extraction gas chromatography-mass spectrometry(HS-SPME-GC-MS)combined with machine learning(ML).It was found that HWPS exhibited a higher viable bacterial count,indicating that consumers preferred FPJ with higher viable bacteria count,while there was no significant differences in color parameters and antioxidant substances between 2 FPJs.A total of 33 volatile compounds were identified in 2 FPJs,respectively.The acid and alcohol levels in HWPS were 37.74％and 32.90％higher,respectively,compared to LWPS,indicating that consumers preferred products with rich volatile compounds.There were 19 key differential volatile compounds screened out by ML.Binary classification models of HWPS and LWPS were established by random forest(RF)and adaptive boosting(AdaBoost)algorithms,and RF algorithm had higher prediction precision and accuracy.According to Shapley Additive exPlanations(SHAP)analysis,the top 9 volatile compounds,including acetic acid,decanoic acid and isopropyl myristate,were the key volatile compounds that affected the scores of HWPS and LWPS.Among them,acetic acid and decanoic acid contributed to positive sensory preferences,while isopropyl myristate had negative sensory effects.KEGG analysis showed that pyruvate metabolism and sulfur metabolism were the main metabolic pathways contributed to formation of volatile compounds.This study used ML combined with SHAP analysis to predict key volatile compounds that influenced sensory preferences,which built a theoretical foundation for using artificial intelligence to aid development of FPJs with typical fermentation flavor and in line with consumer sensory preferences the food industry.关键词
石榴/突尼斯/发酵/机器学习/Shapley可加性特征解释/挥发性风味化合物Key words
pomegranate/Tunisia/fermentation/machine learning(ML)/Shapley Additive exPlanations(SHAP)/volatile compounds分类
轻工纺织引用本文复制引用
邹文惠,潘飞,易俊洁,周林燕..基于机器学习研究乳酸菌发酵石榴汁的风味品质差异形成机制[J].食品工业科技,2025,46(15):36-48,13.基金项目
云南省乡村振兴专项课题(202204BP090025-02) （202204BP090025-02）
云南省千人计划青年人才项目(YNQR-QNRC-2018-102) （YNQR-QNRC-2018-102）
云南省基础研究专项重点基金项目(202401AS070117). （202401AS070117）
