A supervised learning pipeline for decoding nanoparticle antibacterial activity

Antibacterial nanoparticles have significant promise for combating antimicrobial resistance, but their activity is governed by complex interactions between physicochemical properties and bacterial characteristics. Single-parameter studies therefore provide limited guidance for rational nanoparticle design. Here, we report a supervised machine learning framework for decoding nanoparticle antibacterial activity through parameter selection and predictive modeling. We curated datasets from 98 peer-reviewed studies containing experimentally validated minimum inhibitory concentration and minimum bactericidal concentration values. Physicochemical descriptors, including size, shape, surface charge, and synthesis route, were integrated with bacterial characteristics, including Gram stain, morphology, and oxygen requirement, to train regression and classification models. XGBoost and random forest achieved the strongest performance, with R² values of 0.91 and 0.92 for minimum inhibitory concentration prediction and 0.82 and 0.80 for minimum bactericidal concentration prediction, respectively. XGBoost also achieved a classification accuracy of 0.79. Feature analysis identified nanoparticle size, shape, synthesis method, Gram type, and oxygen requirement as key determinants of antibacterial efficacy.