Maintaining the predictive abilities of near-infrared spectroscopy models for the determination of multi-parameters in White Paeony Root

Abstract

Near-infrared spectroscopy (NIRs) has been widely applied in the field of traditional Chinese medicine (TCM). However, the quality of TCM inevitably changes with time and origin, which will result in a decline in the prediction accuracy of constructed NIR models. In this study, models of White Paeony Root were maintained to ensure the accuracy of the predictive values of new samples. The applicability of the original models to the new samples were validated by principal component analysis and model evaluation indexes. Chemometric methods including spectral subtraction correction (SSC) and hierarchical cluster analysis-recalibration (HCA-RC) were introduced to maintain the constructed models. The results demonstrated that, according to the performances of the models, the model maintenance ability of HCA-RC was stronger than that of SSC. With the application of HCA-RC, relative standard error of prediction of moisture, albiflorin and paeoniflorin content were reduced from 5.650%, 29.83% and 39.89% to 2.350%, 14.23% and 14.57%, respectively. This result revealed that HCA-RC could better extract common features from historical and new batches of samples. In addition, the updated HCA-RC models can still be applied for the rapid detection of White Paeony Root. Model maintenance ameliorated the problems of the original models that occurred after a period of application. The updated models contained the characteristics of new batches of samples, which had the potential for long-term continuous application and improved detection efficiency.