Construction of a universal near-infrared quantitative model for carbocysteine tablets and discussion of a training set selection strategy
Objective:To establish a rapid quantitative analysis model for carbocysteine tablets,and to study the number of samples in the training set so as to improve the predictive performance.Methods:A total of 140 batches of carbocysteine tablets samples from 19 national manufacturers were recruited for the study,and the scanning near-infrared spectra of all samples was collected.The total sample set was respectively divided as 120 types,115 types,111 types,106 types,102 types,96 types,91 types,86 types,81 types,76 types,71 types,66 types,61 types,57 types,52 types,47 types and 42 types using clustering analysis.One sample was then randomly selected from each cluster to construct the training set,and the number of samples in training set equaled the number of clusters.Therefore,the optimal calibration models obtained from the automatic optimization routine was implemented in the OPUS Quant 2 software using different numbers of calibration samples.Finally,the best model was chosen and verified.Results:61 samples were selected to establish the quantitative model.The root mean square errors of cross validation and prediction were 1.17％ and 1.08％ for the model for carbocysteine tablets,and correlation coefficient was 0.995;the quantitative model can quickly predict the content of carbocysteine tablets,with active pharmaceutical ingredient (API,)mass fraction ranging from 22.28％ to 85.15％.Conclusion:In this study,the carbocysteine tablets universal quantitative model performances good accuracy and durability.The prediction ability of the established quantitative analysis model was improved by selecting appropriate number of samples.
expectorant drugscarbocysteine tabletsnear-infrared analysis methodsample set of establish modelcalibration settraining setthe number of samplesmodel parametersprediction ability of the modelclustering analysisfast content analysis method
NETL
NSTL
万方数据
维普
