Research On The Establishment Of Near-infrared Qualitative And Quantitative Rapid Detection Model For Honeysuckl

Lonicerae Japonicae FLOS had a high price and a large market demand.It was often mixed with Lonicerae FLOS or replaced with substandard ones by illegal merchants,which posed significant risks in clinical medication safety.In traditional detection methods,the accuracy of macroscopic identification was low,and TLC qualitative identification and HPLC quantification were time-consuming and cumbersome.In recent years,near infrared(NIR)spectroscopy technology,which had been widely studied,was one of the effective methods for achieving qualitative and quantitative rapid detection.Many studies had reported that NIR combined with chemometrics had established quantitative methods for rapid detection of various indicator components in Lonicerae Japonicae FLOS,as well as qualitative methods such as authenticity and origin identification.However,there were still problems such as inconsistent sample pretreatment requirements in different studies and model validation of NIR based on single experiment verification of near Infrared(Sev-NIR)and so on,which made the theoretical performance inconsistent with the practical application and cannot detect multiple indicators at the same time.Therefore,this study compared the effects of different powder particle size and thickness on the NIR spectra of Lonicerae Japonicae FLOS and determined the optimal pretreatment conditions.Based on the Sev-NIR condition,the model was established and the problems of neglected Sev-NIR verification were found under multiple experiment verification of near Infrared(Mev-NIR)condition.The method and requirements for establishing a practical NIR rapid detection model were established.Qualitative and quantitative NIR rapid detection models for Lonicerae Japonicae FLOS were established and verified through random experiment verification of near infrared(Rev-NIR),which proved the prediction ability of Lonicerae Japonicae FLOS NIR rapid detection models in close proximity to actual application scenarios.The main research results were as follows:1.The effects of particle size and thickness on NIR spectra were investigated.When particle size≤250 μm and thickness ≥ 15 mm,the correlation coefficient of NIR spectrum of the same sample was greater than 0.95,the rms value and Euclidean distance were less than 0.020 and 0.680 respectively,and there was no significant change.Its repeatability and stability were acceptable.This condition could ensure the consistency of samples determined by NIR and HPLC.2.The quantitative detection models of chlorogenic acid,phenolic acid and luteoloside were established(HPLC results as reference values)by using partial least square method(PLS).The qualitative identification models of Lonicerae Japonicae FLOS and Lonicerae FLOS were established by using Mahalanobis distance method and PLS discriminant analysis(PLSDA).The models with the highest predictive power under Sev-NIR condition were screened for revalidation at Mev-NIR.The average relative residual and RSD of the predicted values for the same sample under Mev-NIR conditions were much higher than those under Sev-NIR condition.The results of linear regression and paired T-test simultaneously prove that models that only passed the Sev-NIR condition did not achieve theoretical prediction ability in practical applications.By comparing the above two validation results,it was found that factors that ignore Sev-NIR condition and had an impact on the actual prediction ability of the models included:independent research with interference intervals,too low number of modeled spectra,significant digits of input and output values,and lack of contribution of modeled parameters.3.By controlling the consistency of all variables to study the unchanged parts of the results,it was determined that the modeling intervals with basically no interference were 1866～1961 nm,2019～2123 nm,2240～2270 nm,2349～2368 nm,2454～2554 nm.By calculating the correlation between each wavelength and the true content of the indicator components,reference ranges for high correlation of chlorogenic acid.phenolic acid and luteoloside were determined.The minimum requirements for different NIR spectra of the same sample were determined to be 8 and the number of significant digits of input and output values to be 4.4.Contribution of spectral preprocessing methods and modeling intervals,to the performance of the models were investigated.Then the optimal preprocessing methods and modeling intervals for different models were determined.The quantitative detection models of chlorogenic acid,phenolic acid,luteoloside(HPLC results as reference values)and moisture content(the results of the toluene method as a reference value)were establishedby using PLS.PLS-DA was used to establish authenticity identification models for Lonicerae Japonicae FLOS and Lonicerae FLOS,as well as four production area identification models(Julu,Hebei,Pingyi,Shandong,Fengqiu,Henan,Tongwei,Gansu).Validating above modelsunder Mev-NIR and Rev-NIR conditions:(1)The maximum value of R2 between the predicted value and the reference value of chlorogenic acid,phenolic acid,luteoloside quantitative models under the Mev-NIR condition exceeded 0.6(P<0.05),which under the Rev-NIR condition were>0.5(P<0.05).The maximum values of the R2 of the moisture content models were>0.7(P<0.05).After excluding abnormal samples,the average predicted relative residuals of chlorogenic acid,phenolic acid,luteoloside,and moisture content could reach 3.93%,3.27%,9.15%,and 3.17%under Mev-NIR condition,and 6.04%,4.16%,12.14%,and 5.22%under Rev-NIR condition.This indicated that the above models had high prediction ability in practical application scenarios,while the prediction ability for constant components was higher than that for trace components.(2)The best prediction accuracy of Lonicerae Japonicae FLOS and Lonicerae FLOS identification models under Mev-NIR and Rev-NIR conditions was>98%,and the maximum value of R2 was>0.8(P<0.05).The best prediction accuracy of the production area differentiation models of Tongwei,Gansu under Mev-NIR and Rev-NIR conditions was 91.88%and 85.42%.The maximum values of R2 exceeded 0.6 and 0.4(P<0.05).The production area differentiation models of Julu,Hebei could be well distinguished under Mev-NIR condition.The production area differentiation models of Pingyi,Shandong and Fengqiu,Henan had low identification ability under Mev-NIR and Rev-NIR conditions.It indicated that PLS-DA was very suitable for identifying Lonicerae Japonicae FLOS and Lonicerae FLOS,and had high identification ability for Lonicerae Japonicae FLOS from Tongwei,Gansu and Julu,Hebei.