Integrated kinetic modeling with multivariate analysis: New insights into magnesium chloride's strain-specific effects on flavor in radish paocai

Reducing sodium in fermented vegetables via MgCl 2 substitution is a critical public health goal, but it poses significant flavor challenges. This study employed an integrated framework of kinetic modeling and machine learning to investigate the strain-specific effects of MgCl 2 on radish paocai fermented with Levilactobacillus brevis PL6-1 (RL) and Leuconostoc citreum PX3-1 (RX). The fermentation outcomes were critically strain-dependent, with the RL system proving robust and yielding a preferred sensory profile, while the RX system exhibited a biphasic growth response and produced undesirable sulfur compounds. A Multilayer Perceptron (MLP) model, developed to decipher these differences, identified 3-(methylthio)propyl isothiocyanate (for RL) and decanal (for RX) as the key chemical drivers of this divergence. Sensory analysis confirmed the RL group's superiority, with a 10 % MgCl 2 substitution proving a viable strategy specifically for this strain, maintaining acceptability without inducing significant off-notes. Our work concludes that starter culture selection is the primary factor for mitigating the risks of off-flavor development and achieving desirable sensory outcomes in low-sodium fermentation, a challenge effectively deciphered by our integrated modeling approach. • Strain-specific responses dictate success of MgCl 2 substitution in paocai. • L. brevis shows robust fermentation; Leu. citreum risks off-flavor production. • Integrating kinetic modeling and machine learning reveals key flavor drivers. • MLP model pinpoints 3-(methylthio)propyl isothiocyanate and decanal as key flavor drivers. • Starter culture choice is paramount for low-sodium fermentation success.