Making use of available and emerging data to predict the hazards of engineered nanomaterials by means of in silico tools: A critical review

Abstract The heterogeneity of engineered nanomaterials (NMs), with respect to physicochemical properties and observed (eco)toxicological effects, makes their case-by-case testing for Risk Assessment unsustainable in terms of costs, time, and number of test animals. This has required the development of Integrated Approaches to Testing and Assessment (IATA) in compliance with the 3R (Replacement, Reduction, and Refinement) principles of reducing animal testing to assist industries and regulators in decision making related to the safety of NMs. The application of non-testing in silico methods such as (Quantitative) Structure-Activity Relationships ((Q)SARs) and Grouping for Read-Across is fundamental to IATA. Therefore, the goal of this manuscript is to critically review the available in silico methods to predict the hazards for human health posed by NMs by applying the OECD (Q)SAR model validation principles. Moreover, we provide an overview on the existing databases, highlighting the importance of data curation. We find that available approaches are applied mainly to predicting in vitro endpoints, and that most of the models do not build on such internationally recognized databases, but on in-house datasets or on data retrieved from the literature. Our analysis shows that more efforts are needed to properly develop and validate in silico models for hazard prediction as alternatives to in vivo tests. Moreover, the Applicability Domain (highlighting the chemical space where such models were trained and hence provide more reliable results) of such models is not always assessed, and the uncertainty and sensitivity of the current methods are not sufficiently evaluated. To address these challenges, we propose a roadmap for future research in this area, including the adoption of more advanced Machine Learning techniques.