Development of machine learning models to predict the activity of chemical compounds against the obesity-associated melanocortin-4 receptor

Abstract

Monogenic obesity caused by mutations in the Melanocortin-4 Receptor (MC4R) gene remains a significant health challenge, despite numerous efforts to find effective treatments. The MC4R is a promising target for drug development due to its role in energy homeostasis and adipose tissue formation. This thesis explores the hybridization of Machine Learning and Molecular Modeling techniques to identify potential ligands that may act as agonists against the obesity-associated MC4R. This study aimed to develop a predictive model using a dataset of 1,906 chemical compounds and 208 RDKit molecular descriptors to classify molecules based on their activity against MC4R. Additionally, 2,000 natural compounds were evaluated using three molecular docking software platforms to identify potential ligands for human MC4R (hMC4R) based on interaction patterns and binding affinities. The machine learning model was used to predict ligand activity, and their properties were further analyzed using two ADMET tools. The final model demonstrated strong efficiency of activity prediction, achieving 94.28% accuracy and an AUC of 0.98. The molecular docking experiments identified six natural compounds as potential hMC4R ligands, with most sharing the same chemical scaffold. However, the ADMET analysis did not yield accurate or reliable results, limiting the ability to fully assess the safety profiles of the identified compounds. Despite this limitation, the findings suggest that the flavone scaffold could serve as a template for designing novel agonists.