Machine learning-driven nanoparticle toxicity

Abstract

This study presents a comprehensive machine learning-driven analysis to understand and predict the toxicity of nanoparticles (NPs), a crucial aspect in ensuring the safe application of nanotechnology in medicine, pharmaceuticals, biotechnology, and various other industries. By using a robust dataset, we deployed Random Forest (RF) and Light Gradient Boosting Machine (LightGBM) algorithms to identify key NP features that significantly influence cellular toxicity. The integration of Shapley Additive exPlanations (SHAP) values provided an interpretative insight into the predictive models, allowing for a quantitative assessment of feature impact. Our findings highlighted the inverse relationship between NP concentration and cell viability and the heightened toxicity of smaller NPs due to their larger surface-to-volume ratios. Notably, the LightGBM model's sensitivity to zeta potential elucidates the nuanced impact of surface charge on cytotoxic effects. The results from this investigation can guide the synthesis of safer NPs, emphasized the need to consider these critical features to mitigate toxicity while maintaining functional integrity. The study underlines the complexity of NP toxicity modeling and the necessity for advanced analytical methods to capture the multifaceted nature of nanomaterial interactions with biological systems. This work lays the groundwork for future research aimed at refining NP design for safer biomedical applications and consumer products, marking a significant step towards responsible nanotechnology development. • ML models (RF & LightGBM) effectively predict NP toxicity. • NP concentration inversely correlates with cell viability. • Smaller NPs exhibit heightened toxicity due to large surface area. • Zeta potential critically influences NP toxicity predictions. • Findings guide safer NP design for biomedical applications.