Multi-objective optimization of ultrasound-assisted extraction for enhanced bioactive recovery from bee bread

This study aimed to optimize ultrasound-assisted extraction conditions for improving the recovery of phenolic compounds and antiradical activity from bee bread (perga). A three-factor Box-Behnken design comprising ultrasound amplitude (30-70%), extraction time (8-22 min), and ethanol concentration (20-80%, v/v) was combined with Gaussian Process surrogate modeling and a single-cycle GP-based multi-objective Bayesian optimization (GP-MOBO) candidate-selection strategy. Predictive uncertainty and Pareto-based analysis were used to identify a balanced extraction condition, which was subsequently validated experimentally and compared with an NUS and conventional extraction treatments. The selected ultrasound-assisted extraction condition (UAEopt; 56.5% amplitude, 19.0 min, and 56.8% ethanol) yielded a total phenolic content of 17.32 mg GAE/g, compared with 11.76 mg GAE/g in the matched NUS (p < 0.01). Antiradical activity also increased from 3.215 mg TEAC/g in NUS to 4.17 mg TEAC/g in UAEopt (p < 0.05). HPLC-DAD analysis showed that UAEopt produced the highest total concentration of the quantified phenolic compounds (44.19 ± 1.23 μg/g), with gallic acid as the predominant compound. LC-MS analysis further demonstrated that UAEopt yielded the highest total free amino acid concentration among the evaluated extraction conditions (23,409.32 ± 766.26 μg/g), with glutamic acid, aspartic acid, and proline as the major components. These findings demonstrate that Gaussian Process-guided multi-objective optimization can support the selection of effective ultrasound-assisted extraction conditions and improve the recovery of multiple bioactive fractions from bee bread.