Skin sensitisation testing in practice: Applying a stacking meta model to cosmetic ingredients.

Recently, several non-animal approaches contributing to the identification of skin sensitisation hazard have been introduced. Their validation and acceptance has largely been directed towards regulatory classification. Considering the driving force for replacement of in vivo tests centred on cosmetics, it is reasonable to ask how well the new approaches perform in this respect. In the present study, 219 substances, largely cosmetic raw materials (including dyes, preservatives and fragrances), have been evaluated in our Defined Approach integrating a stacking meta model (version 5), incorporating the individual outcomes of 3 in vitro validated methods (Direct Peptide Reactivity Assay, Keratinosens™, U-SENS™), 2 in silico tools (TIMES SS, TOXTREE) and physicochemical parameters (volatility, pH). Stacking meta model outcomes were compared with existing local lymph node assay (LLNA) data. Non-sensitisers comprised 68/219; 86 were weak/moderate and 65 were stronger sensitisers. The model version revision demonstrate the gain to discriminate sensitizers to non-sensitiser when the in silico TIMES model is incorporated as input parameter. The 85% to 91% accuracy for the cosmetics categories, indicates the stacking meta model offers value for the next generation risk assessment framework. These results pinpoint the power of the stacking meta model relying on a confidence based on the probability given in any individual prediction.

Assessment of a defined approach based on a stacking prediction model to identify skin sensitization hazard.

Skin sensitization is an important toxicological endpoint in the safety assessment of chemicals and cosmetic ingredients. Driven by ethical considerations and European Union (EU) legislation, its assessment has progressed from the reliance on traditional animal models to the use of non-animal test methods. It is generally accepted that the assessment of skin sensitization requires the integration of various non-animal test methods in defined approaches (DAs), to cover the mechanistic key events of the adverse outcomes pathway (AOP) (OECD, 2014). Several case studies for DAs predicting skin sensitization hazard or potency have been submitted to the OECD, including a stacking meta-model developed by L'Oréal Research & Innovation (OECD, 2017b; Del Bufalo et al., 2018; Noçairi et al., 2016). The present study evaluated the predictive performance of the defined approach integrating a stacking meta-model incorporating in silico, in chemico and in vitro assays, using the Cosmetics Europe (CE) skin sensitization database. Based on the optimized prediction cut-offs, the defined approach provided a hazard prediction for 97 chemicals with a sensitivity of 91%, a specificity of 76% and accuracy of 86% (kappa of 0.67) against human skin sensitization hazard data and a sensitivity of 85%, specificity of 91% and accuracy of 87% (kappa of 0.67) against Local Lymph Node Assay (LLNA) hazard data. A comparison of the in vivo LLNA with human hazard data for the same 97 chemicals showed a sensitivity of 92%, specificity of 51% and accuracy of 78% (kappa of 0.48). Thus, the defined approach showed a higher degree of concordance, as compared to the LLNA for predicting human skin sensitization hazard. Moreover, a comparison with the six DAs selected for evaluation of their predictivity in the study by Kleinstreuer et al. (2018) showed a similar high accuracy of 86% for 97 overlapping chemicals. The next step will be an independent evaluation of the DA for its integration in the performances based test guidelines (PBTG) for skin sensitization.

Predicting in vivo effect levels for repeat-dose systemic toxicity using chemical, biological, kinetic and study covariates.

In an effort to address a major challenge in chemical safety assessment, alternative approaches for characterizing systemic effect levels, a predictive model was developed. Systemic effect levels were curated from ToxRefDB, HESS-DB and COSMOS-DB from numerous study types totaling 4379 in vivo studies for 1247 chemicals. Observed systemic effects in mammalian models are a complex function of chemical dynamics, kinetics, and inter- and intra-individual variability. To address this complex problem, systemic effect levels were modeled at the study-level by leveraging study covariates (e.g., study type, strain, administration route) in addition to multiple descriptor sets, including chemical (ToxPrint, PaDEL, and Physchem), biological (ToxCast), and kinetic descriptors. Using random forest modeling with cross-validation and external validation procedures, study-level covariates alone accounted for approximately 15% of the variance reducing the root mean squared error (RMSE) from 0.96 log10 to 0.85 log10 mg/kg/day, providing a baseline performance metric (lower expectation of model performance). A consensus model developed using a combination of study-level covariates, chemical, biological, and kinetic descriptors explained a total of 43% of the variance with an RMSE of 0.69 log10 mg/kg/day. A benchmark model (upper expectation of model performance) was also developed with an RMSE of 0.5 log10 mg/kg/day by incorporating study-level covariates and the mean effect level per chemical. To achieve a representative chemical-level prediction, the minimum study-level predicted and observed effect level per chemical were compared reducing the RMSE from 1.0 to 0.73 log10 mg/kg/day, equivalent to 87% of predictions falling within an order-of-magnitude of the observed value. Although biological descriptors did not improve model performance, the final model was enriched for biological descriptors that indicated xenobiotic metabolism gene expression, oxidative stress, and cytotoxicity, demonstrating the importance of accounting for kinetics and non-specific bioactivity in predicting systemic effect levels. Herein, we generated an externally predictive model of systemic effect levels for use as a safety assessment tool and have generated forward predictions for over 30,000 chemicals.

Combination of dynamic time warping and multivariate analysis for the comparison of comprehensive two-dimensional gas chromatograms: application to plant extracts.

Comprehensive two-dimensional gas chromatography (GC x GC) is now recognized as the preferred technique for the detailed analysis and characterization of complex mixtures of volatile compounds. However, for comparison purposes, taking into account all the information contained in the chromatogram is far from trivial. In this paper, it is shown that the combination of peak alignment by dynamic time warping and multivariate analysis facilitated the comparison of complex chromatograms of tobacco extracts. The comparison is shown to be efficient enough to provide a clear discrimination among three types of tobacco. A tentative interpretation of loadings is presented in order to give access to the compounds which differ from one sample to another. Once located, mass spectrometry was used to identify markers of tobacco type.