Sensory score prediction and key aroma compounds characterization in fermented chopped pepper

Abstract

Fermented chopped pepper (FCP) exhibits complex and variable aroma profiles, making it challenging to accurately predict its sensory scores and identify key aroma compounds. In this study, electronic nose (E-nose) combined with machine learning methods were applied for the prediction of FCPs sensory scores. The random forest (RF) demonstrated the highest predictive accuracy among support vector machine (SVM), multiple linear regression (MLR), and back propagation neural network (BPNN). E-nose combined with the trained RF was used to predict the sensory scores of FCPs from eight regions. Totally, 97 volatile compounds and 19 odor-active compounds were detected by GC × GC-O-Q-TOF-MS in the top-performing sample (FCP-1). Among these, 34 compounds exhibited odor activity values (OAV) greater than 1. Aroma recombination and omission experiments confirmed that linalool, phenethyl alcohol, methional, 3-isobutyl-2-methoxypyrazine, ethyl trans-4-decenoate, β-ionone, spiroxide, ethyl 2-methylbutyrate, α-terpineol, 4-ethylphenol, β-damascenone, and nerolidol were the key aroma compounds in FCP-1.