Computational prediction of toxicity has reached new heights as a result of decades of growth in the magnitude and diversity of biological data. Public packages for statistics and machine learning make model creation faster. New theory in machine learning and cheminformatics enables integration of chemical structure, toxicogenomics, simulated and physical data in the prediction of chemical health hazards, and other toxicological information. Our earlier publications have characterized a toxicological dataset of unprecedented scale resulting from the European REACH legislation (Registration Evaluation Authorisation and Restriction of Chemicals). These publications dove into potential use cases for regulatory data and some models for exploiting this data. This article analyzes the options for the iden­tification and categorization of chemicals, moves on to the derivation of descriptive features for chemicals, discusses different kinds of targets modeled in computational toxicology, and ends with a high-level perspective of the algorithms used to create computational toxicology models.

Amid growing efforts to advance the replacement, reduction, and refinement of the use of animals in research, there is a growing recognition that in vitro testing of medical devices can be more effective, both in terms of cost and time, and also more reliable than in vivo testing. Although the technological landscape has evolved rapidly in support of these concepts, regulatory acceptance of alternative testing methods has not kept pace. Despite the acceptance by regulators of some in vitro tests (cytotoxicity, gene toxicity, and some hemocompatibility assays), many toxicity tests still rely on animals (irritation, sensitization, acute toxicity, reproductive/developmental toxicity), even where other industrial sectors have already abandoned them. Bringing about change will require a paradigm shift in current approaches to testing – and a concerted effort to generate better data on risks to human health from exposure to leachable chemicals from medical devices, and to boost confidence in the use of alternative methods to test devices. To help advance these ideas, stir debate about best practices, and coalesce around a roadmap forward, the JHU-Center for Alternatives to Animal Testing (CAAT) hosted a symposium believed to be the first gathering dedicated to the topic of in vitro testing of medical devices. Industry representatives, academics, and regulators in attendance presented evidence to support the unique strengths and challenges associated with the approaches currently in use as well as new methods under development, and drew next steps to push the field forward from their presentations and discussion.

The objective of the present work was to review the availability and predictive value of non-animal toxicokinetic approaches and to evaluate their current use in European risk evaluations of food contaminants, additives and food contact materials, as well as pesticides and medicines. Results revealed little use of quantitative animal or human kinetic data in risk evaluations of food chemicals, compared with pesticides and medicines. Risk evaluations of medicines provided sufficient in vivo kinetic data from different species to evaluate the predictive value of animal kinetic data for humans. These data showed a relatively poor correlation between the in vivo bioavailability in rats and dogs versus that in humans. In contrast, in vitro (human) kinetic data have been demonstrated to provide adequate predictions of the fate of compounds in humans, using appropriate in vitro-in vivo scalers and by integration of in vitro kinetic data with in silico kinetic modelling. Even though in vitro kinetic data were found to be occasionally included within risk eval­uations of food chemicals, particularly results from Caco-2 absorption experiments and in vitro data on gut-microbial conversions, only minor use of in vitro methods for metabolism and quantitative in vitro-in vivo extrapolation methods was identified. Yet, such quantitative predictions are essential in the development of alternatives to animal testing as well as to increase human relevance of toxicological risk evaluations. Future research should aim at further improving and validating quantitative alternative methods for kinetics, thereby increasing regulatory acceptance of non-animal kinetic data.

Chemical hypersensitivity is an immunological response towards foreign substances, commonly referred to as sensitizers, which gives rise primarily to the clinical symptoms known as allergic contact dermatitis. For the purpose of mitigating risks associated with consumer products, chemicals are screened for sensitizing effects. Historically, such predictive screenings have been performed using animal models. However, due to industrial and regulatory demand, animal models for the purpose of sensitization assessment are being replaced by non-animal testing methods, a global trend that is spreading across industries and market segments. To meet this demand, the Genomic Allergen Rapid Detection (GARD) assay was developed. GARD is a novel, cell-based assay that utilizes the innate recognition of xenobiotic substances by dendritic cells, as measured by a multivariate readout of genomic biomarkers. Following cellular stimulation, chemicals are classified as sensitizers or non-sensitizers based on induced transcriptional profiles. Recently, a number of non-animal methods were comparatively evaluated by Cosmetics Europe, using a coherent and blinded test panel of reference chemicals with human and local lymph node assay data, comprising a wide range of sensitizers and non-sensitizers. The outcome of the GARD assay is presented in this paper. It was demonstrated that GARD is a highly functional assay with a predictive performance of 83% in this Cosmetics Europe dataset. The average accumulated predictive accuracy of GARD across independent datasets was 86% for skin sensitization hazard.

Test methods to assess the skin sensitization potential of a substance usually use threshold criteria to dichotomize continuous experimental read-outs into yes/no conclusions. The threshold criteria are prescribed in the respective OECD test guidelines and the conclusion is used for regulatory hazard assessment, i.e., classification and labelling of the substance. We can identify a borderline range (BR) around the classification threshold within which test results are inconclusive due to a test method’s biological and technical variability. We quantified BRs in the prediction models of the non-animal test methods DPRA, LuSens and h-CLAT, and of the animal test LLNA, respectively. Depending on the size of the BR, we found that between 6% and 28% of the substances in the sets tested with these methods were considered borderline. When the results of individual non-animal test methods were combined into integrated testing strategies (ITS), borderline test results of individual tests also affected the overall assessment of the skin sensitization potential of the testing strategy. This was analyzed for the 2-out-of-3 ITS: Four out of 40 substances (10%) were con­sidered borderline. Based on our findings we propose expanding the standard binary classification of substances into “positive”/“negative” or “hazardous”/“non-hazardous” by adding a “borderline” or “inconclusive” alert for cases where test results fall within the borderline range.

Contact allergy induced by certain chemicals is a common health concern, and several alternative methods have been developed to fulfill the requirements of European legislation with regard to hazard assessment of potential skin sensi­tizers. However, validated methods, which provide information about the potency of skin sensitizers, are still lacking. The cell-based assay Genomic Allergen Rapid Detection (GARD), targeting key event 3, dendritic cell activation, of the skin sensitization AOP, uses gene expression profiling and a machine learning approach for the prediction of chemicals as sensitizers or non-sensitizers. Based on the GARD platform, we here expanded the assay to predict three sensitizer potency classes according to the European Classification, Labelling and Packaging (CLP) Regulation, targeting cate­gories 1A (strong), 1B (weak) and no cat (non-sensitizer). Using a random forest approach and 70 training samples, a potential biomarker signature of 52 transcripts was identified. The resulting model could predict an independent test set consisting of 18 chemicals, six from each CLP category and all previously unseen to the model, with an overall accuracy of 78%. Importantly, the model was shown to be conservative and only underestimated the class label of one chemical. Furthermore, an association of defined chemical protein reactivity with distinct biological pathways illustrates that our transcriptional approach can reveal information contributing to the understanding of underlying mechanisms in sensitization.

Skin sensitization is the process by which a substance induces an allergic reaction following skin contact. The process has been described as an adverse outcome pathway (AOP), including several key events, from skin penetration and covalent protein binding, to keratinocyte activation, dendritic cell activation and T-lymphocyte proliferation. The in vitro assay KeratinoSens™ measures the activation of keratinocytes. It is fully accepted at a regulatory level (OECD TG 442d) and complies with a range of legislation including the EU Cosmetics Regulation, REACH, and the CLP Regulation.
Currently, many in vitro methods use animal-derived components in their cell culture systems. Many stakeholders in the cosmetics industry have both scientific and ethical concerns relating to this issue and have stated a strong preference for fully human in vitro test systems. We have adapted the KeratinoSens™ method to animal product-free condi­tions, and carried out an in-house validation with 21 reference substances, including those listed in the Performance Standards associated with OECD TG442d. The modified method was shown to be equivalent to the Validated Reference Method (VRM), with comparable values for accuracy (85.7%), sensitivity (84.6%) and specificity (87.5%), and all acceptance criteria being met. In Europe, data generated by the adapted method may be used in REACH submissions, and we are now seeking approval to list the adaptation in OECD TG 442d, enabling formal compliance with a range of global regulations.