基于指纹图谱的Box-Behnken响应面法结合BP神经网络多指标优化决明子总蒽醌提取工艺

Multi-Index Optimization of Extraction Process for Total Anthraquinone in Cassia Seed by Box-Behnken Response Surface Method Combined with BP Neural Network Based on Fingerprint

目的 基于中药指纹图谱技术,采用Box-Behnken响应面法结合back-propagation(BP)神经网络多指标优化一步提取决明子总游离蒽醌(总蒽醌)并与《中国药典》2020年版方法比较。方法 将决明子直接用稀盐酸回流水解后再同时用二氯甲烷同步回流萃取制备提取物,建立提取物指纹图谱。以总峰面积标准化数值、橙黄决明素与大黄酚的峰面积归一化值及指纹图谱相似度,计算综合评价指标。采用Box-Behnken响应面法考察料液比、水解时间、二氯甲烷用量等因素对综合评价指标的影响,获得最佳提取工艺。建立BP神经网络模型,选取合理数据进行学习和验证并预测决明子的最佳提取工艺。结果 采用Box-Behnken响应面法得到的决明子最佳提取工艺为:料液比为1∶30(g·mL^(-1))、水解时间为30 min、二氯甲烷用量为35 mL;采用BP神经网络预测得到的决明子最佳提取工艺为:料液比为1∶30(g·mL^(-1))、水解时间为45 min、二氯甲烷用量为35 mL。Box-Behnken响应面法、BP神经网络验证结果及2020年版《中国药典》方法的综合评价指标平均值为0.861 3、0.855 1、0.749 7。结论 将Box-Behnken响应面法试验结果与BP神经网络预测结果相互检验,择优确定了决明子总蒽醌的最佳提取工艺,该提取工艺优于2020年版《中国药典》方法。

OBJECTIVE To optimize the one-step extraction method of total free anthraquinones(total anthraquinone)from Cassia seed by Box-Behnken response surface method combined with back-propagation(BP)neural network based on the fingerprint technology of traditional Chinese medicine,and compare with the method of Chinese Pharmacopoeia(2020).METHODS Cassia seed was directly hydrolyzed with dilute hydrochloric acid and then extracted with dichloromethane at the same time to prepare the extract,and the fingerprint of the extract was established.The comprehensive evaluation indexes were calculated by using the standardized value of total peak area,the normalized peak area values of aurantio-obtusin and chrysophanol and the similarity of fingerprint.The effects of solid-liquid ratio,hydrolysis time and dichloromethane dosage on the comprehensive evaluation indexes were investigated by Box-Behnken response surface method and the best extraction process was obtained.BP Neural network model was established,reasonable data was selected to learn and verify,and the best extraction process of Cassia seed was predicted.RESULTS The optimum extraction process of Cassia seed obtained by Box-Behnken response surface method was as follows:the solid-liquid ratio was 1∶30(g·mL^(-1)),the hydrolysis time was 30 min and the amount of dichloromethane was 35 mL.The optimum extraction process of Cassia seed predicted by BP neural network was as follows:the solid-liquid ratio was 1∶30(g·mL^(-1)),the hydrolysis time was 45 min and the amount of dichloromethane was 35 mL.The average values of comprehensive evaluation indexes of Box-Behnken response surface method,BP neural network verification results and the methods of Chinese Pharmacopoeia 2020 edition were 0.8613,0.8551 and 0.7497 respectively.CONCLUSION The test results of Box-Behnken response surface method and the prediction results of BP neural network are tested each other,and the best extraction process of total anthraquinone from Cassia seed is selected,which is better than the method of Chinese Pharmacopoeia 2020.