Discovering new agents active against methicillin-resistant Staphylococcus aureus with ligand-based approaches.

To discover new agents active against methicillin-resistant Staphylococcus aureus (MRSA), in silico models derived from 5451 cell-based anti-MRSA assay data were developed using four machine learning methods, including naïve Bayesian, support vector machine (SVM), recursive partitioning (RP), and k-nearest neighbors (kNN). A total of 876 models have been constructed based on physicochemical descriptors and fingerprints. The overall predictive accuracies of the best models exceeded 80% for both training and test sets. The best model was employed for the virtual screening of anti-MRSA compounds, which were then validated by a cell-based assay using the broth microdilution method with three types of highly resistant MRSA strains (ST239, ST5, and 252). A total of 12 new anti-MRSA agents were confirmed, which had MIC values ranging from 4 to 64 mg/L. This work proves the capacity of combined multiple ligand-based approaches for the discovery of new agents active against MRSA with cell-based assays. We think this work may inspire other lead identification processes when cell-based assay data are available.