RESUMO
Near-infrared spectroscopy (NIRS) combined with chemometrics was used to analyze the main active ingredients including chlorogenic acid, caffeic acid, luteoloside, baicalin, ursodesoxycholic acid, and chenodeoxycholic acid in the Tanreqing injection. In this paper, first, two hundred samples collected in the product line were divided into the calibration set and prediction set, and the reference values were determined by the High Performance Liquid Chromatography- Diode Array Detector/Evaporative Light Scattering Detector (HPLC-DAD/ELSD) method. Partial least squares (PLS) analysis was implemented as a linear method for models calibrated with different preprocessing means. Wavelet transformation (WT) was introduced as a variable selection technique by means of multiscale decomposition, and wavelet coefficients were employed as the input for modeling. Furthermore, two nonlinear approaches, least squares support vector machine (LS-SVM) and Gaussian process (GP), were applied to exploit the complicated relationship between the spectra and active ingredients. The optimal models for each ingredient were obtained by LS-SVM and GP methods. The performance of the final models was evaluated by the root mean square error of calibration (RMSEC), root mean square error of cross-validation (RMSECV), root mean square error of prediction (RMSEP) and correlation coefficient (R). All of the models in the paper give a good calibration ability with an R value above 0.92, and the prediction ability is also satisfactory, with an R value higher than 0.85. The overall results demonstrate that nonlinear models are more stable and predictable than linear ones, and they will be more suitable for the CHM system when high accuracy analysis is required. It can be concluded that NIRS with the LS-SVM and GP modeling methods is promising for the implementation of process analytical technology (PAT) in the pharmaceutical industry of Chinese herbal injections (CHIs).
