Identification of Antibacterial Hits Associated with Penicillin-Binding Protein 2 in Escherichia coli Using a Comprehensive Property Spectrum and Fivefold Maximum Drug-Likeness Strategy

Purpose: Early-stage antibacterial candidate selection requires balancing antibacterial activity with broader developability-related properties. This study developed and applied a Fivefold Maximum Drug-Likeness strategy (5F-MDL) for prioritizing antibacterial candidates against Escherichia coli . Methods: An ensemble of deep learning models generated a 33-dimensional property spectrum covering physicochemical, pharmacokinetic, efficacy-related, safety, and stability endpoints. Approximately 16 million commercial molecules were screened, and fifteen candidates were experimentally evaluated by disk diffusion and broth microdilution. Molecular docking, molecular dynamics simulations, and a Bocillin-FL competition assay examined potential PBP2-associated interactions. Results: The fifteen prioritized candidates showed high property-spectrum similarity to reference antibiotics, with scores ranging from 0.929 to 0.971. Broth microdilution identified several molecules with measurable antibacterial activity, among which M2 showed the most balanced overall profile, including an MIC of 25.6 μg/mL and the largest inhibition zone among the candidates. Docking in the 549 Å 3 active-site-associated cavity of PBP2 supported a plausible M2–PBP2 binding pose, and 200 ns molecular dynamics simulations supported the stability of the modeled complex. The Bocillin-FL assay further suggested that M2 could interfere with probe labeling of PBP2 in vitro. Conclusion: The 5F-MDL workflow provides a multidimensional property-spectrum-based approach for early-stage antibacterial candidate prioritization. M2 was identified as a preliminary lead-like hit, although its mechanism, safety profile, and broader applicability require further validation.