PSO-GA-BPNN optimization, physicochemical characterization, and molecular docking reveal the antioxidant potential of persimmon leaf polysaccharides

Persimmon leaves, as a functional food and traditional Chinese medicine, have long been used in China. However, few studies on the optimal extraction and antioxidant evaluation of polysaccharides from persimmon leaves have been reported. Thus, the optimization of the enzyme-assisted ultrasonic extraction (EAUE) of persimmon leaf polysaccharides (PLPs) was investigated using a hybrid back-propagation neural network combined with a particle swarm optimization and genetic algorithm (PSO-GA-BPNN) model, with the response surface methodology (RSM) as a comparison in this study. The results revealed that the optimal EAUE conditions for the PSO-GA-BPNN were as follows: an enzyme concentration of 50.0 mL/g, ultrasonication temperature of 67 °C, ultrasonication time of 60 min, a liquid-to-material ratio of 20 mL/g, and ultrasonication power of 350 W. Under these parameters, the polysaccharide extraction yield was 12.11%, which was better than the 10.23% yield optimized by RSM. The PLPs were characterized as homogeneous polysaccharides (Mw 126.5 kDa, Mw/Mn 1.23) with a random coil conformation, and were composed mainly of arabinose (31.79%), galactose (30.66%), and galacturonic acid (12.82%). Based on the structural features of the PLPs, five characteristic oligosaccharide fragments were constructed to represent local structural units for molecular docking. Network pharmacology revealed MAPK3, TNF, and IL6 as key oxidative stress targets. All five oligosaccharide fragments could dock into the active pockets of these targets, and their binding regions largely overlapped with those of quercetin. Fragment_A (GalAα1-2Rha) showed binding energies closest to those of quercetin, likely because of electrostatic interactions from its carboxyl group and spatial complementarity from its bent conformation. In the antioxidant activity assays, PLPs showed DPPH and hydroxyl radical scavenging activities. These findings suggest that the PSO-GA-BPNN is effective for optimizing PLP extraction, and that Fragment_A may be a key structural unit.