Application of chemometric algorithms in the high-performance thin-layer chromatography fingerprint of traditional Chinese medicines

Abstract

Background: The use of HPTLC fingerprinting for the analysis of traditional Chinese medicines (TCMs) usually involves several image-processing steps. However, these image-processing steps are time consuming. Objective: We describe a new approach that applies artificial neural networks (ANN) directly to raw high-performance thin-layer chromatography HPTLC images. Methods: This approach combines image processing and chemometric modeling and was used to classify TCMs [dried tangerine eel (Chen Pi), green tangerine peel (Qing Pi), immature bitter orange fruit, and bitter orange fruit (Zhi Qiao)]. Images of the plates were processed with Chempattern and chemometric analysis including PCA, PLS-DA, and kNN were carried out all by ChemPattern. Results: The ANN model has an accuracy of 100.00% in all training, validation, and test sets, indicating excellent predictive performance and good generalization ability. The k-nearest neighbors (kNN) and partial least-square discriminant analysis (PLS-DA) models have accuracies of 90.91 and 72.73%, respectively, with the independent test set. The kNN model is also accurate, simple, and can be easily interpreted. Conclusions: HPTLC fingerprinting, combined with advanced image processing and proper chemometric algorithms, is a simple, efficient, and accurate method for the analysis of TCMs. Highlights: HPTLC fingerprints of four TCM crude drugs derived from Citrus spp. were compared by using image analysis algorithms. A new approach that applied ANN directly to raw HPTLC fingerprint images was described. Three image analysis algorithms based on kNN, PLS-DA and ANN are compared in the paper. The ANN model shows excellent predictive performance with high accuracy in test sets.