Ultrasound-assisted fermentation on aroma and umami enhancement of soybean protein hydrolysates: Machine learning-enhanced flavoromics and molecular insights

Enhancing the flavor of plant-based proteins remains a key challenge in consumer acceptance of alternative food products. This study aimed to investigate the effects of ultrasound-assisted fermentation on the volatile profiles and umami-enhancing effects of soybean protein hydrolysates (SPH). The volatile flavor profiles of SPH subjected to different treatments were comprehensively characterized using GC–MS, GC-IMS, sensory evaluation, and machine learning techniques. GC–MS and GC-IMS analyses showed that ultrasound-assisted fermentation modulated the volatile profile of SPH by increasing pyrazines, esters, and sulfur-containing compounds, while decreasing undesirable aldehydes. Sensory evaluation and electronic tongue results confirmed that fermentation with Lactobacillus fermentum and Pichia fermentans under 150 W ultrasound significantly enhanced aroma complexity and umami intensity. Partial least squares regression combined with machine learning models identified 15 representative aroma compounds that effectively differentiated treatment groups and were closely associated with sensory flavor attributes. Based on relative odor activity value (ROAV) screening and quantitative analysis, ultrasound-assisted co-fermentation was found to facilitate the generation of 11 key aroma-active compounds, such as benzeneacetaldehyde, β-damascenone, 1-octen-3-ol, and dimethyl trisulfide, thereby improving overall aroma quality. Moreover, molecular docking revealed strong interactions between nine aroma compounds and the T1R1/T1R3 umami receptor, with two of them also exhibiting synergistic binding with monosodium glutamate, providing new molecular insights into aroma-umami synergy. Our study supports the application of ultrasound-assisted fermentation as a promising strategy to improve the flavor profiles of plant-based protein ingredients.