ScholarWorks@숙명여자대학교

초록

Objectives: In this study, we newly proposed to apply multiple logistic regression with elastic-net regularization to identify metabolomic markers that predict origins of herbal medicines. Methods: Herbal medicines were collected from two different origins, Korea and China. For each origin, 30 samples were extracted and profiled by nuclear magnetic resonance (NMR) technology. After binning and normalization, we obtained 60 profiles containing density of 240 metabolites. Logistic regression was applied with elastic-net regularization to identify metabolomics markers and build a classifier. We compared the performance of our classifier with the classifier based on orthogonal partial least squares-discriminant analysis (OPLS-DA), which has been commonly used in metabolomics. Results: A total of 14 metabolomic markers were selected to construct the classifier discriminating Korean and Chinese herbal medicines. In predicting the origin of additional samples, our classifier had no misclassification, while the OPLS-DA classifier showed 25% misclassification rate. Conclusions: These results suggest that our method would have advantages against the OPLS-DA, including lower misclassification rate, better interpretability, and no need of an additional analysis for marker identification.

키워드