Optimization of Enzyme-Ultrasound-Assisted Extraction of Almond (Amygdalus communis L.) Phenolics using Response Surface Methodology and Deep Neural Networks, and Their In-Vitro Bioactivities.

This study optimized the extraction of bioactive compounds from almond (Amygdalus communis L.), a medicinal and edible resource from Xinjiang, using ultrasound-assisted enzymatic extraction (pectinase: cellulase). Response surface methodology (RSM) and a deep neural network (DNN) were applied to evaluate ethanol concentration, ultrasound time, solid-liquid ratio, and enzyme dosage on total polyphenol (TPC) and flavonoid content (TFC). Both models predicted extraction outcomes effectively, but RSM (R 2 = 0.9509) outperformed DNN (R 2 = 0.6844) with lower relative error (0.63% vs 0.73%). Optimal RSM conditions were ethanol concentration 65%, ultrasound time 31 min, solid-liquid ratio 29 g/mL, and enzyme dosage 2 mg/g. Under these parameters, guanosine, protocatechuic acid, caffeic acid, narcissin, and taxifolin contents reached 287.78 ± 1.93, 49.01 ± 0.82, 17.35 ± 0.88, 231.41 ± 3.85, and 5.90 ± 0.18 μg/g, respectively. Extracts showed strong antioxidant activity, with maximum DPPH, ABTS, •OH, and O2 •- radical scavenging rates of 87.21%, 74.57%, 66.42%, and 69.03%. They also inhibited α-amylase (80.86%), α-glucosidase (82.27%), and pancreatic lipase (72.73%), suggesting hypoglycemic and lipid-lowering potential. However, concentrations above 250 μg/mL were cytotoxic to RAW264.7 cells. This work offers a feasible strategy for almond bioactive utilization.