Chemoinformatic and machine learning fusion for designing silver nanoparticles with enhanced antimicrobial properties

Predicting the antimicrobial efficacy of newly synthesized nanoparticles remains a complex and resource-intensive task. In this study, we present an interpretable machine learning framework for predicting the antimicrobial activity of silver nanoparticles (AgNPs), utilizing a curated dataset from NanoAntimicrobialDB. The dataset includes physicochemical descriptors and synthesis metadata, which are processed through multi-view feature fusion. We integrate structured numerical features, such as particle size, morphology, and surface charge, with unstructured text features from bacterial strains, synthesis methods, and reagents using TF-IDF embedding and Truncated Singular Value Decomposition (SVD). Six machine learning models-Logistic Regression, Random Forest, XGBoost, LightGBM, CatBoost, and Support Vector Machine (SVM)-were trained and evaluated, with threshold optimization performed using Youden's J statistic. Among the models, CatBoost achieved the highest ROC AUC of 0.9496. At the same time, Logistic Regression achieved the highest accuracy after threshold tuning. SHapley Additive exPlanations (SHAP) analysis identified synthesis method, particle size, nanoparticle stabilization, and surface charge as critical factors influencing antimicrobial efficacy. These results highlight the potential of the proposed framework to predict antimicrobial activity and guide the rational design of next-generation antimicrobial nanoparticles.