The efficient recovery of bioactive compounds is a critical challenge demanding advanced process optimization and deep mechanistic understanding. This study presents a multiscale framework for optimizing ultrasound-assisted extraction (UAE) of polyphenols in scientifically uncharacterized Huangshan–Chaoqing (HC) green tea, used as a model system. Machine learning (ML) approaches, such as Extreme Gradient Boosting (XGboost), Random Forest (RF), and Artificial Neural Network (ANN) and kinetic models were combined with molecular dynamics (MD) simulations to optimize UAE and explore the mechanisms involved. Among the ML models, ANN and XGBoost demonstrated superior accuracy in predicting optimal total phenolic content (TPC) (R2 = 0.97) and DPPH radical scavenging activity (R2 = 0.99), identifying power density as an essential process parameter. Moreover, the predicted optimal conditions were experimentally validated, with an absolute average deviation of 3.92 mgGAE/100g and 0.25 mgTrolox/100g for TPC and DPPH, respectively. Kinetic studies showed that a phenomenological model accurately described the two-stage extraction process involving an initial rapid rate followed by a slower rate until equilibrium. SEM revealed severe cellular disruption and enhanced porosity under UAE. MD simulations driven by experimentally measured bulk temperatures from UAE demonstrated that thermal effects enhance catechin diffusivity by increasing molecular flexibility and reducing intermolecular hydrogen bonds. The diffusion coefficients calculated from the MD simulations agreed with those derived from the experimental kinetic data, validating the computational model. This integrated research provides a multiscale understanding of the UAE process, establishing a rational framework for designing and intensifying the extraction of high-value natural products.
扫码关注我们
求助内容:
应助结果提醒方式:
