-to- Extrapolation on Lung Toxicity Induced by Metal Oxide Nanoparticles Data-Mining.

While analyses are commonly employed for chemical risk assessments, predicting chronic lung toxicity induced by engineered nanoparticles (ENMs) still faces many challenges due to complex interactions at multiple nanobio interfaces. In this study, we developed a rigorous method to compile published evidence on the lung toxicity of metal oxide nanoparticles (MeONPs) and revealed previously overlooked -to- extrapolation (IVIVE) relationships. A comprehensive multidimensional data set containing 1102 data points, 75 pulmonary toxicological biomarkers, and 20 features (covering effects, physicochemical properties, and exposure conditions) was constructed. An IVIVE approach that related effects at the cellular level to lung toxicity in rodent model was established with prediction accuracy reaching 89 and 80% in training and test sets. Experimental validation was conducted by testing chronic lung fibrosis of 8 new MeONPs in 32 independent mice, with prediction accuracy reaching 88%. The IVIVE model indicated that the proinflammatory cytokine IL-1β in THP-1 cells could serve as an marker to predict lung toxicity. The IVIVE model showed great promise for minimizing unnecessary animal tests and understanding toxicological mechanisms.