Non-animal methods to predict skin sensitization (II): an assessment of defined approaches *.

Skin sensitization is a toxicity endpoint of widespread concern, for which the mechanistic understanding and concurrent necessity for non-animal testing approaches have evolved to a critical juncture, with many available options for predicting sensitization without using animals. Cosmetics Europe and the National Toxicology Program Interagency Center for the Evaluation of Alternative Toxicological Methods collaborated to analyze the performance of multiple non-animal data integration approaches for the skin sensitization safety assessment of cosmetics ingredients. The Cosmetics Europe Skin Tolerance Task Force (STTF) collected and generated data on 128 substances in multiple in vitro and in chemico skin sensitization assays selected based on a systematic assessment by the STTF. These assays, together with certain in silico predictions, are key components of various non-animal testing strategies that have been submitted to the Organization for Economic Cooperation and Development as case studies for skin sensitization. Curated murine local lymph node assay (LLNA) and human skin sensitization data were used to evaluate the performance of six defined approaches, comprising eight non-animal testing strategies, for both hazard and potency characterization. Defined approaches examined included consensus methods, artificial neural networks, support vector machine models, Bayesian networks, and decision trees, most of which were reproduced using open source software tools. Multiple non-animal testing strategies incorporating in vitro, in chemico, and in silico inputs demonstrated equivalent or superior performance to the LLNA when compared to both animal and human data for skin sensitization.

Non-animal methods to predict skin sensitization (I): the Cosmetics Europe database.

Cosmetics Europe, the European Trade Association for the cosmetics and personal care industry, is conducting a multi-phase program to develop regulatory accepted, animal-free testing strategies enabling the cosmetics industry to conduct safety assessments. Based on a systematic evaluation of test methods for skin sensitization, five non-animal test methods (DPRA (Direct Peptide Reactivity Assay), KeratinoSensTM, h-CLAT (human cell line activation test), U-SENSTM, SENS-IS) were selected for inclusion in a comprehensive database of 128 substances. Existing data were compiled and completed with newly generated data, the latter amounting to one-third of all data. The database was complemented with human and local lymph node assay (LLNA) reference data, physicochemical properties and use categories, and thoroughly curated. Focused on the availability of human data, the substance selection resulted nevertheless resulted in a high diversity of chemistries in terms of physico-chemical property ranges and use categories. Predictivities of skin sensitization potential and potency, where applicable, were calculated for the LLNA as compared to human data and for the individual test methods compared to both human and LLNA reference data. In addition, various aspects of applicability of the test methods were analyzed. Due to its high level of curation, comprehensiveness, and completeness, we propose our database as a point of reference for the evaluation and development of testing strategies, as done for example in the associated work of Kleinstreuer et al. We encourage the community to use it to meet the challenge of conducting skin sensitization safety assessment without generating new animal data.

Prediction of skin sensitization potency using machine learning approaches.

The replacement of animal use in testing for regulatory classification of skin sensitizers is a priority for US federal agencies that use data from such testing. Machine learning models that classify substances as sensitizers or non-sensitizers without using animal data have been developed and evaluated. Because some regulatory agencies require that sensitizers be further classified into potency categories, we developed statistical models to predict skin sensitization potency for murine local lymph node assay (LLNA) and human outcomes. Input variables for our models included six physicochemical properties and data from three non-animal test methods: direct peptide reactivity assay; human cell line activation test; and KeratinoSens™ assay. Models were built to predict three potency categories using four machine learning approaches and were validated using external test sets and leave-one-out cross-validation. A one-tiered strategy modeled all three categories of response together while a two-tiered strategy modeled sensitizer/non-sensitizer responses and then classified the sensitizers as strong or weak sensitizers. The two-tiered model using the support vector machine with all assay and physicochemical data inputs provided the best performance, yielding accuracy of 88% for prediction of LLNA outcomes (120 substances) and 81% for prediction of human test outcomes (87 substances). The best one-tiered model predicted LLNA outcomes with 78% accuracy and human outcomes with 75% accuracy. By comparison, the LLNA predicts human potency categories with 69% accuracy (60 of 87 substances correctly categorized). These results suggest that computational models using non-animal methods may provide valuable information for assessing skin sensitization potency. Copyright © 2017 John Wiley & Sons, Ltd.

Multivariate models for prediction of human skin sensitization hazard.

One of the Interagency Coordinating Committee on the Validation of Alternative Method's (ICCVAM) top priorities is the development and evaluation of non-animal approaches to identify potential skin sensitizers. The complexity of biological events necessary to produce skin sensitization suggests that no single alternative method will replace the currently accepted animal tests. ICCVAM is evaluating an integrated approach to testing and assessment based on the adverse outcome pathway for skin sensitization that uses machine learning approaches to predict human skin sensitization hazard. We combined data from three in chemico or in vitro assays - the direct peptide reactivity assay (DPRA), human cell line activation test (h-CLAT) and KeratinoSens™ assay - six physicochemical properties and an in silico read-across prediction of skin sensitization hazard into 12 variable groups. The variable groups were evaluated using two machine learning approaches, logistic regression and support vector machine, to predict human skin sensitization hazard. Models were trained on 72 substances and tested on an external set of 24 substances. The six models (three logistic regression and three support vector machine) with the highest accuracy (92%) used: (1) DPRA, h-CLAT and read-across; (2) DPRA, h-CLAT, read-across and KeratinoSens; or (3) DPRA, h-CLAT, read-across, KeratinoSens and log P. The models performed better at predicting human skin sensitization hazard than the murine local lymph node assay (accuracy 88%), any of the alternative methods alone (accuracy 63-79%) or test batteries combining data from the individual methods (accuracy 75%). These results suggest that computational methods are promising tools to identify effectively the potential human skin sensitizers without animal testing. Published 2016. This article has been contributed to by US Government employees and their work is in the public domain in the USA.

In Silico Prediction of Physicochemical Properties of Environmental Chemicals Using Molecular Fingerprints and Machine Learning.

There are little available toxicity data on the vast majority of chemicals in commerce. High-throughput screening (HTS) studies, such as those being carried out by the U.S. Environmental Protection Agency (EPA) ToxCast program in partnership with the federal Tox21 research program, can generate biological data to inform models for predicting potential toxicity. However, physicochemical properties are also needed to model environmental fate and transport, as well as exposure potential. The purpose of the present study was to generate an open-source quantitative structure-property relationship (QSPR) workflow to predict a variety of physicochemical properties that would have cross-platform compatibility to integrate into existing cheminformatics workflows. In this effort, decades-old experimental property data sets available within the EPA EPI Suite were reanalyzed using modern cheminformatics workflows to develop updated QSPR models capable of supplying computationally efficient, open, and transparent HTS property predictions in support of environmental modeling efforts. Models were built using updated EPI Suite data sets for the prediction of six physicochemical properties: octanol-water partition coefficient (logP), water solubility (logS), boiling point (BP), melting point (MP), vapor pressure (logVP), and bioconcentration factor (logBCF). The coefficient of determination (R2) between the estimated values and experimental data for the six predicted properties ranged from 0.826 (MP) to 0.965 (BP), with model performance for five of the six properties exceeding those from the original EPI Suite models. The newly derived models can be employed for rapid estimation of physicochemical properties within an open-source HTS workflow to inform fate and toxicity prediction models of environmental chemicals.

Integrated decision strategies for skin sensitization hazard.

One of the top priorities of the Interagency Coordinating Committee for the Validation of Alternative Methods (ICCVAM) is the identification and evaluation of non-animal alternatives for skin sensitization testing. Although skin sensitization is a complex process, the key biological events of the process have been well characterized in an adverse outcome pathway (AOP) proposed by the Organisation for Economic Co-operation and Development (OECD). Accordingly, ICCVAM is working to develop integrated decision strategies based on the AOP using in vitro, in chemico and in silico information. Data were compiled for 120 substances tested in the murine local lymph node assay (LLNA), direct peptide reactivity assay (DPRA), human cell line activation test (h-CLAT) and KeratinoSens assay. Data for six physicochemical properties, which may affect skin penetration, were also collected, and skin sensitization read-across predictions were performed using OECD QSAR Toolbox. All data were combined into a variety of potential integrated decision strategies to predict LLNA outcomes using a training set of 94 substances and an external test set of 26 substances. Fifty-four models were built using multiple combinations of machine learning approaches and predictor variables. The seven models with the highest accuracy (89-96% for the test set and 96-99% for the training set) for predicting LLNA outcomes used a support vector machine (SVM) approach with different combinations of predictor variables. The performance statistics of the SVM models were higher than any of the non-animal tests alone and higher than simple test battery approaches using these methods. These data suggest that computational approaches are promising tools to effectively integrate data sources to identify potential skin sensitizers without animal testing. Published 2016. This article has been contributed to by US Government employees and their work is in the public domain in the USA.

CERAPP: Collaborative Estrogen Receptor Activity Prediction Project.

BACKGROUND: Humans are exposed to thousands of man-made chemicals in the environment. Some chemicals mimic natural endocrine hormones and, thus, have the potential to be endocrine disruptors. Most of these chemicals have never been tested for their ability to interact with the estrogen receptor (ER). Risk assessors need tools to prioritize chemicals for evaluation in costly in vivo tests, for instance, within the U.S. EPA Endocrine Disruptor Screening Program. OBJECTIVES: We describe a large-scale modeling project called CERAPP (Collaborative Estrogen Receptor Activity Prediction Project) and demonstrate the efficacy of using predictive computational models trained on high-throughput screening data to evaluate thousands of chemicals for ER-related activity and prioritize them for further testing. METHODS: CERAPP combined multiple models developed in collaboration with 17 groups in the United States and Europe to predict ER activity of a common set of 32,464 chemical structures. Quantitative structure-activity relationship models and docking approaches were employed, mostly using a common training set of 1,677 chemical structures provided by the U.S. EPA, to build a total of 40 categorical and 8 continuous models for binding, agonist, and antagonist ER activity. All predictions were evaluated on a set of 7,522 chemicals curated from the literature. To overcome the limitations of single models, a consensus was built by weighting models on scores based on their evaluated accuracies. RESULTS: Individual model scores ranged from 0.69 to 0.85, showing high prediction reliabilities. Out of the 32,464 chemicals, the consensus model predicted 4,001 chemicals (12.3%) as high priority actives and 6,742 potential actives (20.8%) to be considered for further testing. CONCLUSION: This project demonstrated the possibility to screen large libraries of chemicals using a consensus of different in silico approaches. This concept will be applied in future projects related to other end points. CITATION: Mansouri K, Abdelaziz A, Rybacka A, Roncaglioni A, Tropsha A, Varnek A, Zakharov A, Worth A, Richard AM, Grulke CM, Trisciuzzi D, Fourches D, Horvath D, Benfenati E, Muratov E, Wedebye EB, Grisoni F, Mangiatordi GF, Incisivo GM, Hong H, Ng HW, Tetko IV, Balabin I, Kancherla J, Shen J, Burton J, Nicklaus M, Cassotti M, Nikolov NG, Nicolotti O, Andersson PL, Zang Q, Politi R, Beger RD, Todeschini R, Huang R, Farag S, Rosenberg SA, Slavov S, Hu X, Judson RS. 2016. CERAPP: Collaborative Estrogen Receptor Activity Prediction Project. Environ Health Perspect 124:1023-1033; http://dx.doi.org/10.1289/ehp.1510267.

Binary classification of a large collection of environmental chemicals from estrogen receptor assays by quantitative structure-activity relationship and machine learning methods.

There are thousands of environmental chemicals subject to regulatory decisions for endocrine disrupting potential. The ToxCast and Tox21 programs have tested ∼8200 chemicals in a broad screening panel of in vitro high-throughput screening (HTS) assays for estrogen receptor (ER) agonist and antagonist activity. The present work uses this large data set to develop in silico quantitative structure-activity relationship (QSAR) models using machine learning (ML) methods and a novel approach to manage the imbalanced data distribution. Training compounds from the ToxCast project were categorized as active or inactive (binding or nonbinding) classes based on a composite ER Interaction Score derived from a collection of 13 ER in vitro assays. A total of 1537 chemicals from ToxCast were used to derive and optimize the binary classification models while 5073 additional chemicals from the Tox21 project, evaluated in 2 of the 13 in vitro assays, were used to externally validate the model performance. In order to handle the imbalanced distribution of active and inactive chemicals, we developed a cluster-selection strategy to minimize information loss and increase predictive performance and compared this strategy to three currently popular techniques: cost-sensitive learning, oversampling of the minority class, and undersampling of the majority class. QSAR classification models were built to relate the molecular structures of chemicals to their ER activities using linear discriminant analysis (LDA), classification and regression trees (CART), and support vector machines (SVM) with 51 molecular descriptors from QikProp and 4328 bits of structural fingerprints as explanatory variables. A random forest (RF) feature selection method was employed to extract the structural features most relevant to the ER activity. The best model was obtained using SVM in combination with a subset of descriptors identified from a large set via the RF algorithm, which recognized the active and inactive compounds at the accuracies of 76.1% and 82.8% with a total accuracy of 81.6% on the internal test set and 70.8% on the external test set. These results demonstrate that a combination of high-quality experimental data and ML methods can lead to robust models that achieve excellent predictive accuracy, which are potentially useful for facilitating the virtual screening of chemicals for environmental risk assessment.

Identification of heparin samples that contain impurities or contaminants by chemometric pattern recognition analysis of proton NMR spectral data.

Chemometric analysis of a set of one-dimensional (1D) (1)H nuclear magnetic resonance (NMR) spectral data for heparin sodium active pharmaceutical ingredient (API) samples was employed to distinguish USP-grade heparin samples from those containing oversulfated chondroitin sulfate (OSCS) contaminant and/or unacceptable levels of dermatan sulfate (DS) impurity. Three chemometric pattern recognition approaches were implemented: classification and regression tree (CART), artificial neural network (ANN), and support vector machine (SVM). Heparin sodium samples from various manufacturers were analyzed in 2008 and 2009 by 1D (1)H NMR, strong anion-exchange high-performance liquid chromatography, and percent galactosamine in total hexosamine tests. Based on these data, the samples were divided into three groups: Heparin, DS ≤ 1.0% and OSCS = 0%; DS, DS > 1.0% and OSCS = 0%; and OSCS, OSCS > 0% with any content of DS. Three data sets corresponding to different chemical shift regions (1.95-2.20, 3.10-5.70, and 1.95-5.70 ppm) were evaluated. While all three chemometric approaches were able to effectively model the data in the 1.95-2.20 ppm region, SVM was found to substantially outperform CART and ANN for data in the 3.10-5.70 ppm region in terms of classification success rate. A 100% prediction rate was frequently achieved for discrimination between heparin and OSCS samples. The majority of classification errors between heparin and DS involved cases where the DS content was close to the 1.0% DS borderline between the two classes. When these borderline samples were removed, nearly perfect classification results were attained. Satisfactory results were achieved when the resulting models were challenged by test samples containing blends of heparin APIs spiked with non-, partially, or fully oversulfated chondroitin sulfate A, heparan sulfate, or DS at the 1.0%, 5.0%, and 10.0% (w/w) levels. This study demonstrated that the combination of 1D (1)H NMR spectroscopy with multivariate chemometric methods is a nonsubjective, statistics-based approach for heparin quality control and purity assessment that, once standardized, minimizes the need for expert analysts.

Class modeling analysis of heparin 1H NMR spectral data using the soft independent modeling of class analogy and unequal class modeling techniques.

To differentiate heparin samples with varying amounts of dermatan sulfate (DS) impurities and oversulfated chondroitin sulfate (OSCS) contaminants, proton NMR spectral data for heparin sodium active pharmaceutical ingredient samples from different manufacturers were analyzed using multivariate chemometric techniques. A total of 168 samples were divided into three groups: (a) Heparin, [DS] ≤ 1.0% and [OSCS] = 0%; (b) DS, [DS] > 1.0% and [OSCS] = 0%; (c) OSCS, [OSCS] > 0% with any content of DS. The chemometric models were constructed and validated using two well-established methods: soft independent modeling of class analogy (SIMCA) and unequal class modeling (UNEQ). While SIMCA modeling was conducted using the entire set of variables extracted from the NMR spectral data, UNEQ modeling was combined with variable reduction using stepwise linear discriminant analysis to comply with the requirement that the number of samples per class exceed the number of variables in the model by at least 3-fold. Comparison of the results from these two modeling approaches revealed that UNEQ had greater sensitivity (fewer false positives) while SIMCA had greater specificity (fewer false negatives). For Heparin, DS, and OSCS, respectively, the sensitivity was 78% (56/72), 74% (37/50), and 85% (39/46) from SIMCA modeling and 88% (63/72), 90% (45/50), and 91% (42/46) from UNEQ modeling. Importantly, the specificity of both the SIMCA and UNEQ models was 100% (46/46) for Heparin with respect to OSCS; no OSCS-containing sample was misclassified as Heparin. The specificity of the SIMCA model (45/50, or 90%) was superior to that of the UNEQ model (27/50, or 54%) for Heparin with respect to DS samples. However, the overall prediction ability of the UNEQ model (85%) was notably better than that of the SIMCA model (76%) for the Heparin vs DS vs OSCS classes. The models were challenged with blends of heparin spiked with nonsulfated, partially sulfated, or fully oversulfated chondroitin sulfate A, dermatan sulfate, or heparan sulfate at the 1.0, 5.0, and 10.0 wt % levels. The results from the present study indicate that the combination of (1)H NMR spectral data and class modeling techniques (viz., SIMCA and UNEQ) represents a promising strategy for assessing the quality of commercial heparin samples with respect to impurities and contaminants. The methodologies show utility for applications beyond heparin to other complex products.

Combining (1)H NMR spectroscopy and chemometrics to identify heparin samples that may possess dermatan sulfate (DS) impurities or oversulfated chondroitin sulfate (OSCS) contaminants.

Heparin is a naturally produced, heterogeneous compound consisting of variably sulfated and acetylated repeating disaccharide units. The structural complexity of heparin complicates efforts to assess the purity of the compound, especially when differentiating between similar glycosaminoglycans. Recently, heparin sodium contaminated with oversulfated chondroitin sulfate A (OSCS) has been associated with a rapid and acute onset of an anaphylactic reaction. In addition, naturally occurring dermatan sulfate (DS) was found to be present in these and other heparin samples as an impurity due to incomplete purification. The present study was undertaken to determine whether chemometric analysis of these NMR spectral data would be useful for discrimination between USP-grade samples of heparin sodium API and those deemed unacceptable based on their levels of DS, OSCS, or both. Several multivariate chemometric methods for clustering and classification were evaluated; specifically, principal components analysis (PCA), partial least squares discriminant analysis (PLS-DA), linear discriminant analysis (LDA), and the k-nearest-neighbor (kNN) method. Data dimension reduction and variable selection techniques, implemented to avoid over-fitting the training set data, markedly improved the performance of the classification models. Under optimal conditions, a perfect classification (100% success rate) was attained on external test sets for the Heparin vs OSCS model. The predictive rates for the Heparin vs DS, Heparin vs [DS+OSCS], and Heparin vs DS vs OSCS models were 89%, 93%, and 90%, respectively. In most cases, misclassifications can be ascribed to the similarity in NMR chemical shifts of heparin and DS. Among the chemometric methods evaluated in this study, we found that the LDA models were superior to the PLS-DA and kNN models for classification. Taken together, the present results demonstrate the utility of chemometric methods when applied in combination with (1)H NMR spectral analysis for evaluating the quality of heparin APIs.

Determination of galactosamine impurities in heparin samples by multivariate regression analysis of their (1)H NMR spectra.

Heparin, a widely used anticoagulant primarily extracted from animal sources, contains varying amounts of galactosamine impurities. Currently, the United States Pharmacopeia (USP) monograph for heparin purity specifies that the weight percent of galactosamine (%Gal) may not exceed 1%. In the present study, multivariate regression (MVR) analysis of (1)H NMR spectral data obtained from heparin samples was employed to build quantitative models for the prediction of %Gal. MVR analysis was conducted using four separate methods: multiple linear regression, ridge regression, partial least squares regression, and support vector regression (SVR). Genetic algorithms and stepwise selection methods were applied for variable selection. In each case, two separate prediction models were constructed: a global model based on dataset A which contained the full range (0-10%) of galactosamine in the samples and a local model based on the subset dataset B for which the galactosamine level (0-2%) spanned the 1% USP limit. All four regression methods performed equally well for dataset A with low prediction errors under optimal conditions, whereas SVR was clearly superior among the four methods for dataset B. The results from this study show that (1)H NMR spectroscopy, already a USP requirement for the screening of contaminants in heparin, may offer utility as a rapid method for quantitative determination of %Gal in heparin samples when used in conjunction with MVR approaches.