A comprehensive strategy for quality marker discovery using chemical profiling combined with chemometrics, machine learning and network pharmacology analysis: taking Sinomenii Caulis as an example

• Published in: New journal of chemistry Vol. 47; no. 44; pp. 20466 - 20475
• Main Authors: Zhang, Zhiyong, Ren, Mingjun, He, Mulan, Zhu, Yongbo, Huang, Yuming, Qiu, Ping, Hu, Yunfei, Li, Wenlong
• Format: Journal Article
• Language: English
• Published: Cambridge Royal Society of Chemistry 13.11.2023
• Subjects: Accuracy; Algorithms; Alkaloids; Chemometrics; Machine learning; Morphine; Pharmacology; Qualitative analysis; Quality control; Support vector machines
• DOI: 10.1039/D3NJ03669C
• ISSN: 1144-0546

The stems and rhizomes of Sinomenii Caulis (SC) are highly effective in the treatment of rheumatoid arthritis (RA). This study aims to screen potential quality markers using a comprehensive strategy that integrates chemical profiling, chemometrics, machine learning and network pharmacology analysis. First, a database for qualitative analysis of alkaloids in SC was established. Then, non-parametric tests and OPLS-DAs were used to compare the differences between the two medicinal parts, where ROC curves were used to verify the reliability of the results. Subsequently, K-nearest neighbors (KNN), support vector machine (SVM) and random forest (RF) algorithms were employed to calculate the grouping accuracy based on different variables. Finally, network pharmacology was used to analyze the main metabolic pathways of SC in the treatment of RA. A total of 81 alkaloids were identified from SC, including 13 aporphine alkaloids, 18 protoberberine alkaloids, 32 morphine alkaloids, 10 benzylisoquinoline alkaloids, and eight other types of alkaloids. Five compounds were screened by non-parametric tests and OPLS-DAs to differentiate the rhizomes and stems. The machine learning results showed that filtered variables were more responsive to the differences between the two medicinal parts. In addition, the RF model showed a higher classification accuracy than the SVM and KNN models, with an accuracy of 100%. Notably, sinoracutine (23/O), palmatine (41/P), and 8-oxotetrahydropalmatine (81/M) were both active ingredients and differential compounds. This comprehensive strategy may prove to be a powerful technique for screening the quality markers of SC, and can serve as a reference for the design of quality control of other herbs.