Toxicity classification models for metal oxide nanomaterials

Toxicity classification models for metal oxide nanomaterials: Read-across data gap filling and PChem score-based screening approaches


My Kieu Ha, Tung Xuan Trinh, Jang-Sik Choi, Hyung-Gi Byun, Tae Hyun Yoon


My Kieu Ha


College of Natural Sciences, Hanyang University


Metal oxide (MOx) NPs is an important sub-category of engineered nanomaterials (ENMs), as they have very wide usage, such as in cosmetics, textiles, paints and catalysts. However, MOx NPs were also known to induce persistent stress to living organisms, including humans [1]. In risk assessment of ENMs, due to the vast number of ENMs, development of computational methods as a complement to experimental approaches is becoming increasingly important. However, diversity in MOx NPs, heterogeneity of published data format, and lack of standardized measurement/assay protocols are significant obstacles for the development of nanoproperty-toxicity relationship (nano-SAR) models [2].

In this study, we have conducted a comprehensive meta-analysis of 218 published articles on in vitro toxicity of MOx NPs and extracted 7252 cell viability data, accompanied with 14 attributes describing their quantum mechanical, physicochemical, and related experimental conditions. In these comprehensive datasets extracted from meta-analysis, two important challenges, which are quality and completeness of data, have to be overcome for the development of robust nanotoxicity prediction models [3]. To resolve issues in data completeness, we adapted a read-across approach for data gap filling, which involved replacement of the missing values with the data obtained from manufacturers’ specifications. To overcome data quality issue, we proposed a set of PChem-based scoring criteria based on the consistency and reliability of the physicochemical data and measurement protocols and used this criterion set to screen the literature-extracted data and classification nano-SAR models were developed using random forest algorithm.

These two pre-processing steps adapted in this study, 1) read-across approach for data gap filling and 2) PChem score based screening, were found to be effective in improving performances of predictive nano-SAR classification models. We think that these approaches will provide a new framework for taking advantage of the large amount of nanotoxicity data available in the literature for nano-SAR modelling.


1. Colvin, V. L. Nat. Biotechnol., 21, 2003, 1166-1170

2. Oh, E.; Liu, R.; Nel, A.; Gemill, K. B.; Bilal, M.; Cohen, Y.; Medintz, I. L. Nat. Nanotechnol., 11, 2016, 479-486

3. Robinson, R. L. M.; Lynch, I.; Peijnenburg, W.; Rumble, J.; Klaessig, F.; Marquardt, C.; Rauscher, H.; Puzyn, T.; Purian, R.; Aberg, C.; Karcher, S.; Vriens, H.; Hoet, P.; Hoover, M. D.; Hendren, C. O.; Harper, S. L. Nanoscale, 8, 2016, 9919-9943