Machine Learning-Based Prediction of Metal Nanoparticle Toxicity in Human Cell Lines: A Preliminary Contribution to the Development of Nanotoxicology in Peru

Camarena L.; Yauri L.; Romero J.; Tovar N.; Mondalgo J. · Proceedings - 2025 IEEE Latin American Conference on Nanotechnology, LANANO 2025 (2025) · DOI: 10.1109/lanano68466.2025.11512370

Resumen

The increasing use of metallic nanoparticles (NPs) in biomedicine has sparked interest due to their unique nanoscale properties, such as high surface-to-volume ratio and tunable functionalization. However, their potential toxicity in human tissues remains a critical barrier to safe clinical applications. This study presents a preliminary computational nanotoxicology approach using Machine Learning (ML) to predict the toxicity of metallic NPs in human cell lines, providing a foundation for future biomedical research. A dataset integrating physicochemical properties (material type, core size, hydrodynamic size, surface charge, formation enthalpy, among others) and exposure parameters (dose and time) from various cytotoxicity assays was employed. Data were standardized, encoded, and refined through outlier detection and treatment. Feature importance was assessed using ANOVA F-test and Random Forest, revealing that exposure dose (Exposure_dose_µg_per_mL) was the primary predictor of toxicity. To address class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) was applied. Four experimental workflows were designed: two with treated outliers and two with untreated outliers; in each case, one with SMOTE and one without. Six supervised classification models with optimized parameters were evaluated per experiment. Based on AUC, the best model from each case was selected, with Experiment 2 (outliers treated without SMOTE) and the XGBoost model achieving the best overall performance, named RonsocoNano Model, reaching an AUC of 0.963, an accuracy of 0.912, a precision of 0.910, a recall of 0.912, and a F1-score of 0.911. This work represents an initial step toward applying ML in human cell-based nanotoxicology, offering a cost-effective alternative to extensive and heterogeneous in vitro testing. The findings may serve as a foundation for the rational design of less toxic metallic NPs and, in the future, for their evaluation in oncological therapies in Peru, where biomedical nanotechnology remains underdeveloped but holds significant potential.

Instituciones: Universidad Nacional Mayor de San Marcos