Predictive modeling of nanoparticle–bacteria interactions: machine learning approaches for antimicrobial response forecasting

Nanoparticle exposure significantly influences microbial growth, with implications for environmental safety, antimicrobial development, and nanobiotechnology. Predicting bacterial responses experimentally is time-consuming and resource-intensive, necessitating efficient computational alternatives. This study explores the influence of nanoparticle exposure on bacterial growth dynamics, focusing on its relevance to microbiological, environmental, and nanobiotechnology research. A collection of sophisticated machine learning methods including (MLP-ANN), (LSSVM), (DT), (RF), (AdaBoost), and (EL) was employed to construct models predicting bacterial cell concentration (OD600) across various nanoparticle environments. In this work, the learning algorithms were not used in their default configuration; instead, their key hyperparameters were systematically adjusted through the Coupled Simulated Annealing (CSA) metaheuristic to obtain more accurate and stable predictions. Model development was based on a curated experimental database of 2,202 observations, which was randomly split into a training subset of 1,762 records and an independent test subset of 440 records. Each entry in the dataset encoded multiple descriptors of the exposure scenario nanoparticle identity, microbial strain, culture medium, nanoparticle dose, and incubation time so that the resulting models could capture how combinations of physicochemical and biological factors jointly shape microbial growth behavior. The dataset encompassed 12 bacterial and fungal species, such as Cupriavidus necator H16, Escherichia coli, Bacillus cereus, Staphylococcus aureus, and several Aspergillus and Trichoderma species, exposed to silver, cadmium oxide, cerium oxide, iron oxide, and zinc oxide nanoparticles. Sensitivity analysis using Monte Carlo simulations suggested that time exerted the greatest influence on growth predictions, followed by growth medium, nanoparticle concentration, bacterial species, and nanoparticle type. Model performance assessment revealed that the MLP-ANN and EL models consistently achieved the highest predictive precision, as evidenced by superior R² scores, minimal RMSE values, and low AARE percentages across both training and testing datasets. These results highlight potential of machine learning frameworks, especially MLP-ANN and EL, as powerful alternatives to traditional experimental techniques for forecasting microbial growth responses to nanoparticle exposure, thereby reducing laboratory workload and expediting nanotoxicological assessments.