Toward comprehension of the cytotoxicity of heterogeneous TiO2-based engineered nanoparticles: a nano-QSAR approach

Abstract

The current studies demonstrate the potential adverse outcomes of engineered nanoparticles towards humans and the environment. The inadequacy of experimental design and lack of experimental data have blocked the effective toxic assessment for nanoparticles. In this study, two quantitative structure–activity relationship models for nanoparticles (nano-QSAR) have been developed to predict the cytotoxicity of 34 modified TiO₂-based nanoparticles towards Chinese hamster ovary cell line based on the improved SMILES-based optimal descriptors encoded with several available structural properties. Notably, two different code assignment methods, namely hierarchical clustering analysis (HCA) and min–max normalization, were employed to assign alphanumeric codes for physicochemical properties (e.g., particle size, molecular weight, and BET surface area) for comparison purpose. Monte carlo-partial least squares (MC-PLS) was used to develop the nano-QSAR models. The resulting models for both methods showed satisfactory statistical results, with the squared correlation coefficient (R²) values of 0.987 and 0.944 for training sets, and leave-one-out cross-validation parameter (Q_LOO²) of 0.985 and 0.937, respectively. The comparison of the statistical parameters of both models indicated that the HCA method has a better prediction performance than the min–max normalization method. Moreover, both the mole percentage of Ag and BET surface area identified by the nano-QSAR model built with the HCA method are the most influential properties on cytotoxicity of the TiO₂-based nanoparticles. Therefore, the proposed model based on the HCA method could be reasonably expected to provide guidance for selecting safer and more suitable surface modification for the involved metal oxide nanoparticles.