Comprehensive characterization of natural products of Polygonum multiflorum by cheminformatics analysis

Abstract

Introduction: A growing body of evidence has suggested that Polygonum multiflorum Thunb. (PM) may cause PM-herb-induced liver injury (PM-HILI). However, the compounds triggering PM-HILI remain controversial. Herein, we set out to investigate the relationships among natural products of PM (NPPM), drug-induced liver injury (DILI) positive (POS), and negative (NEG) compounds by using cheminformatics methods. Methods: A total of 197 NPPM and 2384 annotated DILI dataset were collected from the literature. Chemical space, physicochemical properties, drug-likeness, intra-set similarity, and scaffold diversity were compared to gain insights into the multiple features of NPPM. An ensemble machine learning (ML) model was constructed to predict the DILI potential of NPPM. Twelve NPPM were selected and tested on HepaRG cells to validate the prediction results. Results: Results of the principal component analysis suggest that NPPM bears more similarity to NEG in terms of chemical space when compared with POS. Besides, NPPM share a moderate overall scaffold diversity and one-third ingredients in NPPM compiled the drug-like rules. The predictive results of ML model show that 28.9% of the small molecules in NPPM bear DILI potential. Further in vitro study by detecting cytotoxicity of representative compounds on HepaRG cells showed that trans- and cis-emodin-physcion dianthrone exhibited the lowest IC50 values of 53.05 µM and 17.11 µM, respectively. Conclusion: ML methods and further validation recognize the liver toxicity of dianthonres and emodins. A proportion of NPPM components exhibit drug-likeness profiles and might offer complementary resources for drug discovery. The results demonstrated machine learning-powered DILI prediction as a useful tool to study potential DILI risk of compounds, providing a basis for further identification of toxins or leads in PM. The codes used and generated in this study are freely available at https://github.com/dreadlesss/Polygonum_database_analysis.

Keywords

• Polygonum multiflorum
• Drug-induced liver injury
• Natural products
• Cheminformatics
• Machine learning