Optimization of extraction and purification processes of six flavonoid components from Radix Astragali using BP neural network combined with particle swarm optimization and genetic algorithm

Radix Astragali (RA), as food and traditional Chinese medicine, has been used for a long history in China. However, few studies on the extraction and purification of the active components from RA have been reported. Thus, the optimal extraction and purification conditions of six flavonoid components (calycosin, calycosin-7-glucoside, formononetin, ononin, 7,2′-dihydroxy-3′,4′- dimethoxyisoflav −7-O-glucoside, and 9,10-dimethoxyptercarpan-3-O-β-D-glucoside) from RA were investigated in this study. BP neural network combined with particle swarm optimization and genetic algorithm (PSO-GA-BPNN) was used to analyze the relations between the extraction/purification conditions and the response value, and search the optimal extraction/purification conditions. Response surface methodology (RSM) was performed simultaneously as a comparison. Furthermore, the DPPH·radical scavenging activity and hydroxyl radical scavenging activity of crude extract and purified extract were detected. A three-layer PSO-GA-BPNN containing 3 input neurons, 3 hidden layer neurons and 1 output neuron was established to investigate the relationship between the extraction/purification conditions and the response value. For extraction, the optimal condition by PSO-GA-BPNN was as follow: ethanol concentration 54%, extraction time 254 min, and solid-to-liquid ratio 18 mL/g. Using the optimal condition, the yield of six components was 2.084 mg/g, which was greater than 1.915 mg/g optimized by RSM. For purification, among nine types of macroporou resins, CAD-40 with higher adsorption ratio and desorption ratio was selected to purify the extract. The optimal adsorption condition by PSO-GA-BPNN was as follow: pH 6.9, adsorption flow rate 1.6 mL/min and solution concentration 0.22 g/mL, while the optimal desorption condition by PSO-GA-BPNN was as follow: ethanol concentration 55%, desorption flow rate 1.5 mL/min and elution volume 8.5 BV. Under the optimal condition, the adsorption capacity and desorption capacity were 1.465 mg/g and 1.259 mg/g, respectively, which were better than 1.381 mg/g and 1.156 mg/g optimized by RSM. These optimal values obtained by PSO-GA-BPNN were significant increased compared with these by RSM according statistical analysis. The antioxidant test showed that the purified extract has better antioxidant activity. These results indicate that PSO-GA-BPNN model was more appropriate and has better indicative value for extraction than RSM, while the same as for purification. Consequently, PSO-GA-BPNN model combined with MPR purification enhanced the antioxidant potential of the RA from this medicinal species.