Nanotechnology products have long since made their way to markets around the world increasing the concerns about whether nanomaterials pose a risk to our environment or health. It has been suggested that engineered nanomaterial (ENM) with broad applications and rapid commercialization need better risk assessment and regulation. However, the refinement of regulations to deal with ENMs is limited by the time consuming and costly nature of in vivo and in vitro toxicity testing. In silico methods offer an inexpensive and rapid mechanism to integrate data from in vitro and in vivo testing and to ultimately predict their toxicity without the need for toxicological evaluations. Quantitative structure activity relationships (QSARs) can be developed to correlate descriptors of chemical compounds with their biological activities to inform risk assessments. As one of the most widely used additives in
paints, sunscreens and electronic devices, zinc oxide nanoparticles (NP) are expected to increase in our environment. Some computational models have been established for simple bare metal NPs; however, none to date have focused on surface modified ZnO NPs. The goal of this project was to use NP toxic response data and determine if the inherent NP surface modification has a predictable effect on toxicity. To assess for hazardous effects caused by ZnO NPs, embryonic zebrafish were selected as vertebrate test species as their transparent tissues allow for easy visual assessment of multiple developmental malformations and their short life span allows for rapid assessments. The physicochemical properties of NP surface modifications were calculated with consideration of fish water pH and electrolyte concentrations. Principal component analysis (PCA) and ordinary kriging (OK) methods were applied to develop our model. To test our model for prediction of more complicated ZnO NPs, we selected 2 additional ZnO NPs that were doped with Fe₂O₃ or Al₂O₃, and determined if they matched our toxicity estimations. Based on this strategy, ENM toxicity could be rapidly estimated from label information and wide range of kriging maps with increasing support from our publically available knowledgebase and global collaborations.