Virtual Screening Based on Machine Learning Explores Mangrove Natural Products as KRAS^G12C Inhibitors

Abstract

Mangrove secondary metabolites have many unique biological activities. We identified lead compounds among them that might target KRASG12C. KRAS is considered to be closely related to various cancers. A variety of novel small molecules that directly target KRAS are being developed, including covalent allosteric inhibitors for KRASG12C mutant, protein–protein interaction inhibitors that bind in the switch I/II pocket or the A59 site, and GTP-competitive inhibitors targeting the nucleotide-binding site. To identify a candidate pool of mangrove secondary metabolic natural products, we tested various machine learning algorithms and selected random forest as a model for predicting the targeting activity of compounds. Lead compounds were then subjected to virtual screening and covalent docking, integrated absorption, distribution, metabolism and excretion (ADME) testing, and structure-based pharmacophore model validation to select the most suitable compounds. Finally, we performed molecular dynamics simulations to verify the binding mode of the lead compound to KRASG12C. The lazypredict function package was initially used, and the Accuracy score and F1 score of the random forest algorithm exceeded 60%, which can be considered to carry a strong ability to distinguish the data. Four marine natural products were obtained through machine learning identification and covalent docking screening. Compound 44 and compound 14 were selected for further validation after ADME and toxicity studies, and pharmacophore analysis indicated that they had a favorable pharmacodynamic profile. Comparison with the positive control showed that they stabilized switch I and switch II, and like MRTX849, retained a novel binding mechanism at the molecular level. Molecular dynamics analysis showed that they maintained a stable conformation with the target protein, so compound 44 and compound 14 may be effective inhibitors of the G12C mutant. These findings reveal that the mangrove-derived secondary metabolite compound 44 and compound 14 might be potential therapeutic agents for KRASG12C.

Keywords

• mangrove natural products
• KRAS^G12C
• machine learning
• molecular docking
• drug discovery
• virtual screening