Modelling the toxicity of a large set of metal and metal oxide nanoparticles using the OCHEM platform

Abstract

Inorganic nanomaterials have become one of the new areas of modern knowledge and technology and have already found an increasing number of applications. However, some nanoparticles show toxicity to living organisms, and can potentially have a negative influence on environmental ecosystems. While toxicity can be determined experimentally, such studies are time consuming and costly. Computational toxicology can provide an alternative approach and there is a need to develop methods to reliably assess Quantitative Structure-Property Relationships for nanomaterials (nano-QSPRs). Importantly, development of such models requires careful collection and curation of data. This article overviews freely available nano-QSPR models, which were developed using the Online Chemical Modeling Environment (OCHEM). Multiple data on toxicity of nanoparticles to different living organisms were collected from the literature and uploaded in the OCHEM database. The main characteristics of nanoparticles such as chemical composition of nanoparticles, average particle size, shape, surface charge and information about the biological test species were used as descriptors for developing QSPR models. QSPR methodologies used Random Forests (WEKA-RF), k-Nearest Neighbors and Associative Neural Networks. The predictive ability of the models was tested through cross-validation, giving cross-validated coefficients q = 0.58-0.80 for regression models and balanced accuracies of 65-88% for classification models. These results matched the predictions for the test sets used to develop the models. The proposed nano-QSPR models and uploaded data are freely available online at http://ochem.eu/article/103451 and can be used for estimation of toxicity of new and emerging nanoparticles at the early stages of nanomaterial development.

Practices and Trends of Machine Learning Application in Nanotoxicology

“…The optimal k value is selected using distances (generally Euclidian distances) as weighting factors for voting, which characterizes compounds' dissimilarity in a multidimensional feature space [76]. The k value can be selected by a cross-validation method [102]. Fourches et al [76] used an algorithm combining kNN and a variable selection procedure to maximize model accuracy.…”

“…The DTF uses data rows left out to validate the model without the requirement of a separate data set. Kovalishyn et al [102] built an ensemble of backpropagation neural networks while applying the kNN method to determine the local correction of the Associative Neural Networks (ASNN). Their ASNN ensemble included 100 networks.…”

“…Since there are a number of X-residuals, a summary is required and this is accomplished by the standard deviation of the X of the matrix corresponding row. Kovalishyn et al [102] used the ensemble predictions standard deviation (STD), which correlates with predictions' accuracy. The method shows that the prediction is more likely to be unreliable if dissimilar models give significantly dissimilar predictions for a case and STD is preferably used as a model uncertainty estimator.…”

Nano-(Q)SAR for Cytotoxicity Prediction of Engineered Nanomaterials

“…The analysis of relative descriptor importance for the built neural network model identified dose, formation enthalpy, exposure time, and hydrodynamic size as the four most important descriptors [57]. However, the advantage of regression models for the analysis of toxicity of NPs was shown in comparison with the classification models on metal NPs and metal oxide NPs [58]: Regression models allow not only qualitative, but also a quantitative evaluation of the studied nanomaterials.…”

“…In conclusion, we have summarized the data relating to 39 QSAR and SAR models (18 for E. coli, nine for HaCaT cells, six for transformed bronchial epithelial cells (BEAS-2B), four for murine myeloid cells (RAW 264.7), and two for rat L2 lung epithelial cells and rat lung alveolar macrophages). Of the 39 models, 12 were built using the MLR method, which is reasonable since MLR has certain advantages compared to classification models [58]. Most of the descriptors in the described models relate to physico-chemical, constitutional, topological, and quantum mechanical types.…”