Estudio in silico mediante modelado QSAR e inteligencia artificial para la predicción de la actividad antimicrobiana de derivados del bencimidazol frente a Staphylococcus aureus

Abstract

Antimicrobial resistance is a global public health problem associated with an increase in infections caused by multidrug-resistant bacteria and a decrease in the effectiveness of available antibiotics. Staphylococcus aureus has been reported as one of the main pathogens in hospitals and communities, with a high prevalence of resistant strains. In Ecuador, studies show a growing resistance of S. aureus to methicillin, implying the need to explore new therapies. In this context, QSAR modeling and artificial intelligence emerge as computational approaches to study the relationship between the chemical structure and biological activity of antimicrobial compounds. The objective of this study was to analyze, in silico, the structure-activity relationship of benzimidazole ring derivatives selected as the central focus of the study due to their recurrent presence in molecules with antimicrobial activity against Staphylococcus aureus. A matrix of 100 compounds with the benzimidazole core and experimental minimum inhibitory concentration (MIC) values was developed from public databases and scientific literature. These structures were geometrically optimized, and molecular descriptors were calculated using PaDEL-Descriptor, constructing a refined matrix of 636 descriptors. QSAR models were developed using ElasticNetCV, Random Forest, and Support Vector Regression, applying cross-validation, residual analysis, applicability domain assessment, and external validation using independent compounds. The results suggest that the nonlinear models adequately reproduced descriptor-activity patterns, while the linear model identified descriptors associated with conformational flexibility, polar surface area, hydrophilicity, and molecular mass, which are related to increases in MIC values. The study shows that the integrated use of QSAR modeling and artificial intelligence allows the analysis and prediction of the antimicrobial activity of benzimidazole derivatives, providing useful structural criteria for future stages of computational research.