Abstract
The invasive plant Heracleum sosnowskyi Manden. is a valuable source of a number of bioactive metabolites that can be used in the pharmaceutical industry and medicine and may have some other applications as well. Today, there is a need to summarize data on these substances as well as analyze the toxicological profile of the metabolites of H. sosnowskyi. In this study, we collected a dataset of 225 metabolites of H. sosnowskyi from different literature sources and performed cluster analysis of their chemical structures; we revealed five main clusters of compounds: terpenoids, aromatic compounds, polyaromatic compounds, fatty acids, and furanocoumarins. In order to fill the gaps in the experimental data on the toxicity of the studied substances, we used machine learning (ML) algorithms previously designed for high-accuracy prediction of toxicity end-points. The ML-based approach allowed us to fill in up to 90% of the missing median lethal dose LD50 (mouse) data for the studied molecules. The validity of each predicted value was confirmed by analyzing the applicability domain of the used ML models. For the calculations and ML modeling, we used the Syntelly chemoinformatics platform. For the most toxic compounds-hydroxycoumarins and furanocoumarins of H. sosnowskyi-the values for hepatotoxicity, drug-induced liver injury (DILI), cardiotoxicity, and carcinogenicity were predicted. Based on the analysis of LD50 values for the mouse animal model, the greatest toxicity for furanocoumarins is expected with the intravenous route of administration (62-450 mg/kg), which can cause drug-induced liver injury. At the same time, the data do not show high cardiotoxicity risks for the studied furanocoumarins. Based on the presented results, we discuss prospects of using some of the compounds as pharmaceutical agents.