Machine learning ensemble models for predicting the antibacterial efficacy of gold nanoparticles
Priya Mary, Abdulhassan Mujeeb · Materials Research Express · 2025
Abstract Antimicrobial resistance (AMR) has been increasing rapidly, emerging as a major global health challenge. Gold nanoparticles (AuNPs) are promising antibacterial agents due to their biocompatibility, low toxicity, and ease of functionalization. However, optimizing the physico-chemical properties of AuNPs and experimental parameters to enhance their antibacterial properties is resource-intensive, time-consuming, and often involves trial and error methods. To address these challenges scientifically, this work adopts Machine learning (ML) techniques to predict the Zone Of Inhibition (ZOI) as antibacterial efficacy of gold nanoparticles against multiple species of bacteria. Exploratory data analysis identifies relationships between dataset features and actual ZOI outcomes. Using the actuals, various ML regression models are trained and then validated for their performance. Ensemble methods, notably the Voting Regressor utilizing three base estimator models, such as eXtreme gradient Boost (XGBoost), Light Gradient Boosting Machine (LightGBM), and CatBoost, demonstrated superior accuracy in predicting antibacterial efficacy. Besides this, ensemble models captured a substantial proportion of the variance in the dependent variable and learned from the underlying complex and non-linear features. Furthermore, SHapley Additive exPlanations (SHAP) analysis was performed as an Explainable Artificial Intelligence (XAI) technique. This revealed further insights into each sample’s prediction by giving individual feature’s marginal contribution. Although core size, dose and shape are the predominant features determining the ZOI prediction, based on individual samples, the marginal contribution can vary significantly across samples. This research streamlines AuNP antibacterial studies by reducing experimental redundancy, saving time and resources, and elucidating factors influencing antibacterial efficacy. It emphasizes the potential of meta-models, like the Voting Regressor, to enhance predictive accuracy.