Dynamic QSAR modeling for predicting in vivo genotoxicity and inflammation induced by nanoparticles and advanced materials: a time-dose-property/response approach

Predicting the health risks of nanoparticles (NPs) and advanced materials (AdMa) is a critical challenge. Due to the complexity and time-consuming nature of experimental testing, there is a reliance on in silico methods such as quantitative structure-activity relationship (QSAR), which, while effective, often fail to capture the dynamic nature of material activity over time-essential for reliable risk assessment. This study develops dynamic QSAR models using machine learning to predict toxicological responses, such as inflammation and genotoxicity, following pulmonary exposure to 39 AdMa across various post-exposure time points and dose levels. By incorporating exposure time, administered dose, and material properties as independent variables, we successfully developed time-dose-property/response models capable of predicting AdMa-induced in vivo genotoxicity in bronchoalveolar lavage fluid cells, lung and liver tissue, and inflammation in terms of neutrophil influx into the lungs of mice. Key factors driving AdMa-induced toxicity were identified, including exposure dose, post-exposure duration time, aspect ratio, surface area, reactive oxygen species generation, and metal ion release. The time-dose-property/response modeling paradigm presented here provides a practical and robust approach for predicting in vivo genotoxicity and inflammation and supports the comprehensive risk assessment of morphologically diverse AdMa.