Machine-learning-guided discovery of defect-engineered ZnO nanostructures for enhanced antibacterial activity

Transition-metal-doped zinc oxide nanoparticles have emerged as promising antibacterial agents owing to their tunable electronic structure and defect chemistry. While numerous experimental studies have reported enhanced antibacterial activity in doped ZnO nanostructures, quantitative frameworks that link physicochemical descriptors to biological response remain limited. Here we reanalyze experimental datasets of ZnO nanoparticles with varying size, dopant composition, and surface modification using a machine-learning approach to uncover the key parameters governing antibacterial behavior. Using structured datasets derived from antibacterial assays including colony-forming unit counts, minimum inhibitory concentration, and zone-of-inhibition measurements, we trained ensemble learning models to predict bacterial growth inhibition as a function of nanoparticle descriptors. Random-forest regression achieved robust predictive performance and enabled quantitative ranking of feature importance. Explainable artificial intelligence using SHAP (Shapley additive explanations) revealed that nanoparticle concentration and defect-mediated electronic structure are the dominant factors controlling antibacterial efficacy, followed by particle size and dopant chemistry. The analysis further demonstrates that Co-doped ZnO exhibits enhanced bactericidal activity compared with Fe-doped and silica-embedded ZnO, consistent with increased oxygen vacancy formation and improved generation of reactive oxygen species under illumination. These findings establish a data-driven framework for linking nanomaterial design parameters with antibacterial performance and provide predictive guidelines for engineering defect-controlled antimicrobial nanomaterials.