Predicting antibacterial activity of silver nanoparticles using physicochemical descriptors.

Quantitative prediction of antibacterial activity in green-synthesized silver nanoparticles (AgNPs) is critical for developing effective and sustainable antimicrobial agents, particularly against multidrug-resistant bacteria. However, variability in synthesis protocols, characterization methods, and biological assays, along with reliance on low-throughput microscopy, limits the ability to systematically link nanoparticle properties to antibacterial performance. Here, we present a materials-informatics framework that predicts inhibition zone diameter using machine-learning models trained on literature-derived datasets. Two complementary descriptor sets were evaluated: structural features from electron microscopy and optical parameters from UV-vis spectroscopy. Ensemble learning models achieved high predictive accuracy, with XGBoost reaching R2 > 0.94 for microscopy-based descriptors and CatBoost achieving R2 ≈ 0.93 for optical features. Feature analysis identified nanoparticle size and bacterial concentration as dominant predictors. These findings demonstrate that UV-vis-derived descriptors can provide predictive capability comparable to microscopy-based features, offering a faster, scalable, and sustainable approach for designing and screening antibacterial nanomaterials.