ИСТИНА

Nowadays, nanomaterials are highly integrated into our daily life. Categorization of the environmental hazards associated with engineered nanomaterials is important for evaluating the potential risks brought by commercialized ENMs. Such a task is so far severely hindered because of insufficient amount of available toxicity data. As biological assays are costly and time-consuming and also face the ethical issue of animal use, computational modeling such as (quantitative) nanostructure-activity relationships (nano-(Q)SARs) is valued as a potential tool to fill in the data gaps. The goal of the present study is evaluate toxicity of Ag nanoparticles for Danio rerio embryos and develop efficient QSAR models that allow predicting the toxicological properties and effects of inorganic nanomaterials (metals and oxides) by using the Online Chemical Modeling Environment (OCHEM). Numerical data on toxicity of nanoparticles to different organisms have been taken from the literature and experiment, and then data have been uploaded in the OCHEM database. The main characteristics of nanoparticles such as chemical composition of nanoparticles, average particle size, shape and information about the biological test species were used as obligatory condition for all properties in OCHEM. We evaluated the toxicity of suspensions of Ag nanoparticles with different stabilizers and different shapes to the embryos. We investigate influence of different type of stabilizers on physicochemical properties, stability and behavior of silver nanoparticles. The harmful behavior of flat and spherical silver NPs to Danio rerio embryos was assessed. We analyzed influence NPs on survival, hatching rate and morphogenesis of Danio rerio embryos. Toxicity of NPs were compared with data for silver ions. Within our study, QSAR models were compared by following the same procedure with different combinations of descriptors and machine learning methods. QSAR methodologies included Random Forests (WEKA-RF), kNearest Neighbors and Associative Neural Networks. The predictive ability of the models was tested through leave-one-out cross-validation, giving a q2=0.69-0.79 for regression models and total accuracies Ac=76- 100% for classification models. Predictions for the external evaluation sets obtained accuracies in the range of 78-100% (for low/high toxicity classifications) and q2=0.70-0.79 for regressions. The method showed itself to be a potential tool for estimation of toxicity of new nanoparticles at early stages of nanomaterial development.