A holistic strategy for the in-depth discrimination and authentication of 16 citrus herbs and associated commercial products based on machine learning techniques and non-targeted metabolomics

Citrus-derived raw medicinal materials are frequently used for health care, flavoring, and therapeutic purposes. However, Due to similarities in origin or appearance, citrus herbs are often misused in the market, necessitating effective differentiation methods. For the first time, this study constructed automated discrimination models for 16 citrus species (239 batches) while previous studies focused on a limited number of species. Seven machine learning models -Tree, Discriminant, Support Vector Machine, K-Nearest Neighbor, Ensemble, Neural Network, and Partial least squares discriminant analysis-were compared, with the Ensemble model achieving 100% accuracy in the test set. 16 Orthogonal partial least squares discriminant analysis models were constructed to screen and identify 53 differential markers. These markers were successfully utilized to determine the absence or presence of specified components in the 20 citrus products. This study provides a comprehensive solution for the quality control of citrus herbs, enabling the differentiation of raw herbs and processed slices, as well as the identification of complex systems such as Chinese patent medicines.