The development and application of ENM promises many benefits to both society and the global economy. However, to ensure responsible development of this emerging technology, governance must be put into place to ensure that any potential risk posed by ENM is fully understood and controlled.

The regulatory landscape is constantly evolving for all substances and products. However, in the case of ENM, challenges are greater because on the nanoscale, properties of a material relevant to its safety and effectiveness may differ from those on the bulk scale.

To aid the responsible development of nanotechnologies, this work package aims to develop an “ENM safety classifier” – a computational based predictive principle for the assessment of ENM safety. Based upon the resulting design and data requirements of the model, specification for a high throughput system for future testing and analysis of further refined or new ENMs will also be produced.

Outcomes of the workpackage 

This work package has developed the ENM Safety Classifier by expanding further the concept and design and implementing it as a usable computational tool; through the prototyping and testing of the tool; and, based upon the resulting design and data requirements of the tool, it has developed a specification for a high throughput system for future testing and analysis of further and new ENMs. The further development of a formal conceptual model and design for the tool by the partners in this WP depended upon access to large volumes of data and information being readily retrieved and abstracted from the data repository in WP11, as well as to other knowledge from preceding WPs. It related particularly to the data management and analysis carried out in WP11, and the involvement in WP11 of IOM (as WP12 leader) – and similarly the Finnish Institute of Occupational Health (FIOH) as the tool developers – has greatly assisted the data availability and its flow.

The ENM Safety Classifier is a computational based predictive tool for the assessment of ENM safety. The tool will facilitate the prediction of hazard classification based on a minimal core set of input data, and is intended to be widely applicable by industry, (in particular to SMEs) regulators and academics involved in the development of materials. This task is linked with Task 11.1 (WP11). Development of the underlying data algorithms was done in WP11, based on analysis of extensive data collected in the project. These algorithms were linked to the minimal core dataset (identified in WP11) with the classification outputs. Key subtasks achieved within this task include (i) the development a specification of the optimum output classes which were achieved with an options analysis in consultation with the stakeholder groups. Options range was from a simple yes/no to consideration of dose for different endpoints, boundaries, uncertainties and the level of information which should be supplied by users. A stakeholder consultation/workshop was arranged to refine this specification. (ii)  The development of a specification of the optimum inputs based on interaction with WP11. In this we identified the kinds of input data were necessary, such as, physical/chemical/morphological data, toxicological and ecotoxicological endpoints. Based on this specification, a data interface was developed to link the output data of WP11 and the input data for the Classifier. Based on these decisions, the algorithm which translated these inputs into safety classifications was implemented. This process has identified the optimal set of data which need to be collected and the format of the data. (iii) The Classifier was calibrated with existing complete data sets containing all of the required data inputs in addition to definitive measures of toxicity.

The Classifier was validated using data from FP7 project MARINA, using the silver nanoparticle, after consultation with industrial partners holding relevant, existing or under-development (industrially focused data sets). The validation of the Classifier consisted of collecting sorting and cleaning the data, running the evaluations, assessment and calibration of the validity of outputs. The Classifier’s classification for silver nanoparticle as a material of ‘medium’ toxicity was accurate as this is well understood in nanosafety research. The Classifier functioning and the process of model validation were demonstrated in a workshop to the expert team from within the project stakeholder group.

A two-phased high-throughput testing platform was suggested to couple diverse technologies and assays with the NANOSOLUTIONS Classifier to group ENMs into low, medium or high toxicity materials. Data produced from cell culture model assays enabled initial and rapid generation of diverse dose-response and condition-dependent measures of the cytotoxicity and genotoxicity of ENMs. Scoring related to effective dose and detectable effects in the cell models enables a first level of “toxic potency” ranking of the ENMs. At selected exposure levels based on this ranking, quantitative PCR-based assessment of seven mRNA species predictive of cytotoxic and genotoxic potency thereafter enables completion of the toxicity classification via a second set of experiments. The selected mRNA species were chosen from the minimal feature set identified by the Classifier on the basis of providing better algorithm-based grouping of the ENMs than sequencing-derived mRNAs, miRNAs, proteins, corona proteins, and intrinsic ENM physiochemical data, and finally, from the ease of rapidly and reproducibly analyzing mRNA species quantitatively. The overall grouping assessment strategy of the platform thereby includes three high-throughput angles to the toxicity classification: initial dose-response cytotoxicity and genotoxicity analyses are coupled thereafter with expression analysis of mRNAs demonstrated by bioinformatics methodology to capture outcome of the cytotoxicity and genotoxicity effects, including separately and in combination. Applying more of the vast data generated in the project under machine learning principles enables further development of the Classifier as it becomes applied to independent EMN safety testing data in the future. The suggested platform would serve to effectively classify a corresponding number of ENMs as those assessed over four years in the NANOSOLUTIONS project within a time frame of 2-3 months, based overall on the generation of potentially 104-105 data points at high throughput.