基于近红外光谱技术的腰痹通胶囊中间体质量分析通用模型建模方法研究  

Study on the modeling method of general model of Yaobitong capsule intermediates quality analysis based on near infrared spectroscopy

作　　者：斯乐婷 张欣 张永超 张江艳 王钧 陈勇[3] 刘雪松[3] 吴永江[3] SI Le-ting;ZHANG Xin;ZHANG Yong-chao;ZHANG Jiang-yan;WANG Jun;CHEN Yong;LIU Xue-song;WU Yong-jiang(Department of Pharmacy,Women's Hospital School of Medicine Zhejiang University,Hangzhou 310006,China;State Key Laboratory on Technologies for Chinese Medicine Pharmaceutical Process Control and Intelligent Manufacture(Jiangsu Kanion Pharmaceutical Co.,Ltd.),Lianyungang 222047,China;College of Pharmaceutical Sciences,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]浙江大学医学院附属妇产科医院药剂科,浙江杭州310006 [2]中药制药过程控制与智能制造技术全国重点实验室(江苏康缘药业股份有限公司),江苏连云港222047 [3]浙江大学药学院,浙江杭州310058

出　　处：《药学学报》2025年第2期471-478,共8页Acta Pharmaceutica Sinica

摘　　要：采用近红外光谱技术(near infrared spectroscopy,NIRS)结合化学计量学建立腰痹通胶囊生产过程中间体的质量分析通用模型,实现三七皂苷R1、人参皂苷Rg1、人参皂苷Re、人参皂苷Rb1、人参皂苷Rd及水分6个关键质量指标的快速检测。以喷干细粉和总混颗粒为研究对象,采用高效液相色谱法测定5种皂苷成分含量、烘干法测定水分含量作为参考值,并采集近红外光谱。经蒙特卡洛交叉验证(Monte Carlo cross validation,MCCV)剔除异常样本后,使用蒙特卡洛-无信息变量消除(Monte Carlo uninformative variables elimination,MC-UVE)、竞争性自适应重加权采样(competitive adaptive reweighted sampling,CARS)选择特征变量,分别采用偏最小二乘回归(partial least squares regression,PLSR)、极限学习机(extreme learning machine,ELM)和蚁狮算法优化的最小二乘向量机(ant lion optimization least squares support vector machine,ALO-LSSVM)建立定量模型,比较模型效果。结果表明,经CARS筛选变量后建立的ALO-LSSVM模型效果最佳,6种指标成分的模型相关系数均大于0.93,相对标准误差控制在6%之内,ALO-LSSVM更适用于数量多、信息丰富的样本,模型预测效果和稳定性得到显著提高。本研究建立的通用模型具有良好的预测效果,可用于腰痹通胶囊中间体的快速检测。The general models for intermediates quality analysis in the production process of Yaobitong capsule were established by near infrared spectroscopy(NIRS)combined with chemometrics,realizing the rapid determination of notoginsenoside R1,ginsenoside Rg1,ginsenoside Re,ginsenoside Rb1,ginsenoside Rd and moisture.The spray-dried fine powder and total mixed granule were selected as research objects.The contents of five saponins were determined by high performance liquid chromatography and the moisture content was determined by drying method.The measured contents were used as reference values.Meanwhile,NIR spectra were collected.After removing abnormal samples by Monte Carlo cross validation(MCCV),Monte Carlo uninformative variables elimination(MC-UVE)and competitive adaptive reweighted sampling(CARS)were used to select feature variables respectively.Based on the feature variables,quantitative models were established by partial least squares regression(PLSR),extreme learning machine(ELM)and ant lion optimization least squares support vector machine(ALO-LSSVM).The results showed that CARS-ALO-LSSVM model had the optimum effect.The correlation coefficients of the six index components were greater than 0.93,and the relative standard errors were controlled within 6%.ALO-LSSVM was more suitable for a large number of samples with rich information,and the prediction effect and stability of the model were significantly improved.The general models with good predicting effect can be used for the rapid quality determination of Yaobitong capsule intermediates.

关 键 词：腰痹通胶囊 中间体 近红外光谱技术 通用模型 蚁狮算法优化的最小二乘向量机 

分 类 号：R917[医药卫生—药物分析学]