Investigating nanotoxicity: uncovering associations and predictive factors through machine learning analysis of published literature

Abstract

Nanotoxicity has become a major concern of human health due to the extensive applications of nanomaterials in several fields. This study investigates nanotoxicity by combining association rule mining and supervised machine learning to overcome their limitations when used independently. The data were collected from published literature, which included nanomaterial properties, experimental protocols, and toxicity outcomes. Association rule mining is employed to identify significant associations and hidden patterns. Meanwhile, supervised learning algorithms offer predictive power towards unseen data. As a result, the XGBoost model demonstrates the highest accuracy, reaching approximately 90%. The analysis of feature importance suggests that toxicity is significantly influenced by coat/functional and material. Concurrently, rule mining and classification machine learning results reveal that testing protocols hold equivalent significance to material traits regarding their impact on toxicity. This allows us to gain deeper insights and understanding of nanotoxicity and its influencing factors, facilitating the development of nanoparticle designs, regulations, and standards that promote the safe use and disposal of nanomaterials.