Machine learning-assisted optimization of ultrasound–ohmic processing of hawthorn vinegar: antidiabetic activity, phenolic profiling, and molecular docking insights

Abstract

Hawthorn vinegar is a fermented product with functional properties, containing phenolic compounds and bioactive ingredients that could promote health. In this study, ultrasound–ohmic (USOH) processing conditions were optimized using a hybrid machine-learning-based approach to maximize the α-glucosidase and α-amylase and inhibitory activities of hawthorn vinegar. A Box–Behnken experimental design with 27 runs was used, including four independent variables: ultrasound amplitude (40–80%), ultrasound duration (2–6 min), ohmic field strength (20–40 V/cm), and ohmic heating time (2–6 min). Thirteen machine learning algorithms were comparatively evaluated using systematic hyperparameter optimization with GridSearchCV and 5-fold cross-validation. The Lasso Poly2 model showed the highest predictive performance for both α-glucosidase and α-amylase inhibition, with CV R2 values of 0.9301 and 0.9299, respectively, and low MAPE values (<1%). Metaheuristic optimization algorithms, including Particle Swarm Optimization (PSO), Differential Evolution (DE), and Gray Wolf Optimization (GWO), converged to similar optimum processing conditions, indicating the robustness of the optimized process region. Under the combined optimal conditions, the experimental α-amylase and α-glucosidase inhibition activities were 39.27 ± 1.36% and 37.54 ± 0.53%, respectively. In addition, USOH treatment significantly enhanced the phenolic profile of hawthorn vinegar compared to thermally pasteurized and untreated samples. In particular, the contents of chlorogenic acid, catechin hydrate, caffeic acid, rutin, naringin, resveratrol, and quercetin were markedly increased after treatment. Additionally, five phenolic compounds were evaluated by molecular docking analysis against α-amylase and α-glucosidase, and the strongest binding affinities were observed for naringin (−7.40 kcal/mol) and chlorogenic acid (−7.17 kcal/mol), respectively. These findings demonstrate that machine learning-assisted ultrasound–ohmic processing can effectively improve the antidiabetic and functional properties of hawthorn vinegar.