INTRODUCTION

Sasang Constitutional Medicine (SCM) is a type of traditional Korean medicine where patients are classified as one of four Sasang constitution types (Sasang type) and medications consisting of medicinal herbs are prescribed according to the Sasang type. Despite the importance of personalized medicine, the operation mechanism is largely unknown. To gain a better understanding, we investigated the compound information that composes Sasang type-specific personalized herbal medicines on both multivariate and univariate levels.

METHODS

Five machine learning classifiers including extremely randomized trees (ERT) were trained to investigate whether the Sasang type can be explained by compound information at the multivariate level. Hierarchical clustering was conducted to determine whether compounds are processed distributedly or specifically. Taxonomic and biosynthetic analyses were conducted on these compounds. A univariate level statistical test was conducted to provide more robust Sasang type-specific compound information.

RESULTS

Using the trained ERT classifier, sixty important compounds were extracted. The sixty compounds were clustered into three groups, corresponding to each Sasang type-prominent compounds, suggesting that most compounds have specific preference for the Sasang type. Structural and biosynthetic characteristics of these Sasang type-prominent compounds were determined based on taxonomy and pathway analyses. Fourteen compounds showed statistically significant relevance with the Sasang type. Additionally, we predicted the Sasang type of unknown herbs, which were confirmed by their biological effects in functional assays.

CONCLUSION

This study investigated the personalized herbal medicines of the SCM using compound information. This study provided information on the chemical characteristics of the compounds that are essential for classifying the Sasang type of medicinal herbs, as well as predictions regarding the Sasang type of the commonly used but unidentified medicinal herbs.