Computational approaches for QSAR-based prediction of MRSA antimicrobial activity

Methicillin-resistant Staphylococcus aureus (MRSA) is a leading cause of nosocomial and community-acquired infections, with an estimated global death rate of 10 million per year by 2050. This study aims to identify effective therapeutic candidates by developing several QSAR models. First, data were downloaded from the ChEMBL34 database, then cleaned and prepared. Next, compound properties were assessed using Lipinski's rule of five, and scaffolds were later analyzed to identify key structural frameworks associated with anti-MRSA activity. After that, various molecular descriptors capturing physicochemical and electronic properties were calculated. Several machine learning algorithms, including LGBM, XGB, SVR and random forest, were used to build the predictive models. Their performance was evaluated using metrics such as R2, Q2, Q2LMO, Q2F3 and CCC, with results indicating high predictive power (all metrics above the thresholds limits). This study highlights the effectiveness of QSAR modelling as a robust approach for discovering new anti-MRSA agents.