Quantum Mechanical Structure−Activity Relationship Analyses for Skin Sensitization

Transfer learning for predicting human skin sensitizers

Computational prioritization of chemicals for potential skin sensitization risks plays essential roles in the risk assessment of environmental chemicals and drug development. Given the huge number of chemicals for testing, computational methods enable the fast identification of high-risk chemicals for experimental validation and design of safer alternatives. However, the development of robust prediction model requires a large dataset of tested chemicals that is usually not available for most toxicological endpoints, especially for human data. A small training dataset makes the development of effective models difficult with insufficient coverage and accuracy. In this study, an ensemble tree-based multitask learning method was developed incorporating three relevant tasks in the well-defined adverse outcome pathway (AOP) of skin sensitization to transfer shared knowledge to the major task of human sensitizers. The results show both largely improved coverage and accuracy compared with three state-of-the-art methods. A user-friendly prediction server was available at https://cwtung.kmu.edu.tw/skinsensdb/predict . As AOPs for various toxicity endpoints are being actively developed, the proposed method can be applied to develop prediction models for other endpoints.

QSAR models of human data can enrich or replace LLNA testing for human skin sensitization

Skin sensitization is a major environmental and occupational health hazard. Although many chemicals have been evaluated in humans, there have been no efforts to model these data to date. We have compiled, curated, analyzed, and compared the available human and LLNA data. Using these data, we have developed reliable computational models and applied them for virtual screening of chemical libraries to identify putative skin sensitizers. The overall concordance between murine LLNA and human skin sensitization responses for a set of 135 unique chemicals was low (R = 28-43%), although several chemical classes had high concordance. We have succeeded to develop predictive QSAR models of all available human data with the external correct classification rate of 71%. A consensus model integrating concordant QSAR predictions and LLNA results afforded a higher CCR of 82% but at the expense of the reduced external dataset coverage (52%). We used the developed QSAR models for virtual screening of CosIng database and identified 1061 putative skin sensitizers; for seventeen of these compounds, we found published evidence of their skin sensitization effects. Models reported herein provide more accurate alternative to LLNA testing for human skin sensitization assessment across diverse chemical data. In addition, they can also be used to guide the structural optimization of toxic compounds to reduce their skin sensitization potential.

Predicting chemically-induced skin reactions. Part I: QSAR models of skin sensitization and their application to identify potentially hazardous compounds

Repetitive exposure to a chemical agent can induce an immune reaction in inherently susceptible individuals that leads to skin sensitization. Although many chemicals have been reported as skin sensitizers, there have been very few rigorously validated QSAR models with defined applicability domains (AD) that were developed using a large group of chemically diverse compounds. In this study, we have aimed to compile, curate, and integrate the largest publicly available dataset related to chemically-induced skin sensitization, use this data to generate rigorously validated and QSAR models for skin sensitization, and employ these models as a virtual screening tool for identifying putative sensitizers among environmental chemicals. We followed best practices for model building and validation implemented with our predictive QSAR workflow using Random Forest modeling technique in combination with SiRMS and Dragon descriptors. The Correct Classification Rate (CCR) for QSAR models discriminating sensitizers from non-sensitizers was 71-88% when evaluated on several external validation sets, within a broad AD, with positive (for sensitizers) and negative (for non-sensitizers) predicted rates of 85% and 79% respectively. When compared to the skin sensitization module included in the OECD QSAR Toolbox as well as to the skin sensitization model in publicly available VEGA software, our models showed a significantly higher prediction accuracy for the same sets of external compounds as evaluated by Positive Predicted Rate, Negative Predicted Rate, and CCR. These models were applied to identify putative chemical hazards in the Scorecard database of possible skin or sense organ toxicants as primary candidates for experimental validation.

Modeling skin sensitization potential of mechanistically hard-to-be-classified aniline and phenol compounds with quantum mechanistic properties

Background

Advanced structure-activity relationship (SAR) modeling can be used as an alternative tool for identification of skin sensitizers and in improvement of the medical diagnosis and more effective practical measures to reduce the causative chemical exposures. It can also circumvent ethical concern of using animals in toxicological tests, and reduce time and cost. Compounds with aniline or phenol moieties represent two large classes of frequently skin sensitizing chemicals but exhibiting very variable, and difficult to predict, potency. The mechanisms of action are not well-understood.

Methods

A group of mechanistically hard-to-be-classified aniline and phenol chemicals were collected. An in silico model was established by statistical analysis of quantum descriptors for the determination of the relationship between their chemical structures and skin sensitization potential. The sensitization mechanisms were investigated based on the features of the established model. Then the model was utilized to analyze a subset of FDA approved drugs containing aniline and/or phenol groups for prediction of their skin sensitization potential.

Results and discussion

A linear discriminant model using the energy of the highest occupied molecular orbital (ϵ_HOMO) as the descriptor yielded high prediction accuracy. The contribution of ϵ_HOMO as a major determinant may suggest that autoxidation or free radical binding could be involved. The model was further applied to predict allergic potential of a subset of FDA approved drugs containing aniline and/or phenol moiety. The predictions imply that similar mechanisms (autoxidation or free radical binding) may also play a role in the skin sensitization caused by these drugs.

Conclusions

An accurate and simple quantum mechanistic model has been developed to predict the skin sensitization potential of mechanistically hard-to-be-classified aniline and phenol chemicals. The model could be useful for the skin sensitization potential predictions of a subset of FDA approved drugs.

Electronic supplementary material

The online version of this article (doi:10.1186/2050-6511-15-76) contains supplementary material, which is available to authorized users.

Prediction of skin sensitization potential using D-optimal design and GA-kNN classification methods

Modelling of skin sensitization data of 255 diverse compounds and 450 calculated descriptors was performed to develop global predictive classification models that are applicable to whole chemical space. With this aim, we employed two automated procedures, (a) D-optimal design to select optimal members of the training and test sets and (b) k-Nearest Neighbour classification (kNN) method along with Genetic Algorithms (GA-kNN Classification) to select significant and independent descriptors in order to build the models. This methodology helped us to derive multiple models, M1-M5, that are stable and robust. The best among them, model M1 (CCR(train) = 84.3%, CCR(test) = 87.2% and CCR(ext) = 80.4%), is based on six neighbours and nine descriptors and further suggests that: (a) it is stable and robust and performs better than the reported models in literature, and (b) the combination of D-optimal design and GA-kNN classification approach is a very promising approach. Consensus prediction based on the models M1-M5 improved the CCR of training, test and external validation datasets by 3.8%, 4.45% and 3.85%, respectively, over M1. From the analysis of the physical meaning of the selected descriptors, it is inferred that the skin sensitization potential of small organic compounds can be accurately predicted using calculated descriptors that code for the following fundamental properties: (i) lipophilicity, (ii) atomic polarizability, (iii) shape, (iii) electrostatic interactions, and (iv) chemical reactivity.

Determinants of skin sensitisation potential

Skin sensitisation is an important toxicological endpoint. The possibility that chemicals used in the workplace and in consumer products might cause skin sensitisation is a major concern for individuals, for employers and for marketing. In European REACH (Registration, Evaluation, and Authorisation of Chemicals) legislation, the sensitising potential should therefore be assessed for chemicals below the 10 ton threshold. Development of methods for prediction of skin sensitisation potential without animal testing has been an active research area for some time, but has received further impetus with the advent of REACH and the EU Cosmetics Directive (EU 2003). This paper addresses the issue of non-animal based prediction of sensitisation by a mechanistic approach. It is known that the sequence of molecular, biomolecular and cellular events between exposure to a skin sensitiser and development of the sensitised state involves several stages, in particular penetration through the stratum corneum, covalent binding to carrier protein, migration of Langerhans cells, presentation of the antigen to naïve T-cells. In this paper each of these stages is considered with respect to the extent to which it is dependent on the chemical properties of the sensitiser. The evidence suggests that, although penetration of the stratum corneum, stimulation of migration and maturation of Langerhans cells, and antigen recognition are important events in the induction of sensitisation, except in certain specific circumstances they can be taken for granted. They are not important factors in determining whether a compound will be a sensitiser or not, nor are they important factors in determining how potent one sensitiser will be relative to another. The ability to bind covalently to carrier protein is the major structure-dependent determinant of skin sensitisation potential. A chemistry-based prediction strategy is proposed involving reaction mechanistic domain assignment, reactivity and hydrophobicity determination, and application of quantitative mechanistic modelling (QMM) or read-across.

Identification of mechanisms of toxic action for skin sensitisation using a SMARTS pattern based approach

Skin sensitisation is a key endpoint under REACH as it is costly and its assessment currently has a high dependency on animal testing. In order to reduce both the cost and the numbers of animals tested, it is likely that (quantitative) structure-activity relationships ((Q)SAR) and read-across methods will be utilised as part of intelligent testing strategies. The majority of skin sensitisers elicit their effect via covalent bond formation with skin proteins. These reactions have been understood in terms of well defined nucleophilic-electrophilic reaction chemistry. Thus, a first step in (Q)SAR analysis is the assignment of a chemical's potential mechanism of action enabling it to be placed in an appropriate reactivity domain. The aim of this study was to design a series of SMARTS patterns capable of defining these reactivity domains. This was carried out using a large database of local lymph node assay (LLNA) results that had had potential mechanisms of action assigned to them using expert knowledge. A simple algorithm was written enabling the SMARTS patterns to be used to screen a database of SMILES strings. The SMARTS patterns were then evaluated using a second, smaller, test set of LLNA results which had also had potential mechanisms of action assigned by experts. The results showed that the SMARTS patterns provided an excellent method of identifying potential electrophilic mechanisms. The findings are supported, in part, by molecular orbital calculations which confirm assignment of reactive mechanism of action. The ability to define a chemical's potential reaction mechanism is likely to be of significant benefit to regulators and risk assessors as it enables category formation and subsequent read-across to be performed.

Allergic contact dermatitis--formation, structural requirements, and reactivity of skin sensitizers

Contact allergy is caused by a wide range of chemicals after skin contact. Its clinical manifestation, allergic contact dermatitis (ACD), is developed upon repeated contact with the allergen. This perspective focuses on two areas that have yielded new useful information during the last 20 years: (i) structure-activity relationship (SAR) studies of contact allergy based on the concept of hapten-protein binding and (ii) mechanistic investigations regarding activation of nonsensitizing compounds to contact allergens by air oxidation or skin metabolism. The second area is more thoroughly reviewed since the full picture has previously not been published. Prediction of the sensitizing capacity of a chemical is important to avoid outbreaks of ACD in the population. Much research has been devoted to the development of in vitro and in silico predictive testing methods. Today, no method exists that is sensitive enough to detect weak allergens and that is robust enough to be used for routine screening. To cause sensitization, a chemical must bind to macromolecules (proteins) in the skin. Expert systems containing information about the relationship between the chemical structure and the ability of chemicals to haptenate proteins are available. However, few designed SAR studies based on mechanistic investigations of prohaptens have been published. Many compounds are not allergenic themselves but are activated in the skin (e.g., metabolically) or before skin contact (e.g., via air oxidation) to form skin sensitizers. Thus, more basic research is needed on the chemical reactions involved in the antigen formation and the immunological mechanisms. The clinical importance of air oxidation to activate nonallergenic compounds has been demonstrated. Oxidized fragrance terpenes, in contrast to the pure terpenes, gave positive patch test reactions in consecutive dermatitis patients as frequently as the most common standard allergens. This shows the importance of using compounds to which people are exposed when screening for ACD in dermatology clinics.

An evaluation of selected global (Q)SARs/expert systems for the prediction of skin sensitisation potential

Skin sensitisation potential is an endpoint that needs to be assessed within the framework of existing and forthcoming legislation. At present, skin sensitisation hazard is normally identified using in vivo test methods, the favoured approach being the local lymph node assay (LLNA). This method can also provide a measure of relative skin sensitising potency which is essential for assessing and managing human health risks. One potential alternative approach to skin sensitisation hazard identification is the use of (Quantitative) structure activity relationships ((Q)SARs) coupled with appropriate documentation and performance characteristics. This represents a major challenge. Current thinking is that (Q)SARs might best be employed as part of a battery of approaches that collectively provide information on skin sensitisation hazard. A number of (Q)SARs and expert systems have been developed and are described in the literature. Here we focus on three models (TOPKAT, Derek for Windows and TOPS-MODE), and evaluate their performance against a recently published dataset of 211 chemicals. The current strengths and limitations of one of these models is highlighted, together with modifications that could be made to improve its performance. Of the models/expert systems evaluated, none performed sufficiently well to act as a standalone tool for hazard identification.

TIMES-SS—A promising tool for the assessment of skin sensitization hazard. A characterization with respect to the OECD validation principles for (Q)SARs and an external evaluation for predictivity

The TImes MEtabolism Simulator platform used for predicting Skin Sensitization (TIMES-SS) is a hybrid expert system that was developed at Bourgas University using funding and data from a Consortium comprising industry and regulators. The model was developed with the aim of minimizing animal testing and to be scientifically valid in accordance with the OECD principles for (Q)SAR validation. TIMES-SS encodes structure-toxicity and structure-skin metabolism relationships through a number of transformations, some of which are underpinned by mechanistic 3D QSARs. Here, we describe the extent to which the five OECD principles are met and in particular the results from an external evaluation exercise that was recently carried out. As part of this exercise, data were generated for 40 new chemicals in the murine local lymph node assay (LLNA) and then compared with predictions made by TIMES-SS. The results were promising with an overall good concordance (83%) between experimental and predicted values. Further evaluation of these results highlighted certain inconsistencies which were rationalized by a consideration of reaction chemistry principles for sensitization. Improvements for TIMES-SS were proposed where appropriate. TIMES-SS is a promising tool to aid in the evaluation of skin sensitization hazard under legislative programs such as REACH.

Global (Q)SARs for skin sensitisation: assessment against OECD principles

As part of a European Chemicals Bureau contract relating to the evaluation of (Q)SARs for toxicological endpoints of regulatory importance, we have reviewed and analysed (Q)SARs for skin sensitisation. Here we consider some recently published global (Q)SAR approaches against the OECD principles and present re-analysis of the data. Our analyses indicate that "statistical" (Q)SARs which aim to be global in their applicability tend to be insufficiently robust mechanistically, leading to an unacceptably high failure rate. Our conclusions are that, for skin sensitisation, the mechanistic chemistry is very important and consequently the best non-animal approach currently applicable to predict skin sensitisation potential is with the help of an expert system. This would assign compounds into mechanistic applicability domains and apply mechanism-based (Q)SARs specific for those domains and, very importantly, recognise when a compound is outside its range of competence. In such situations, it would call for human expert input supported by experimental chemistry studies as necessary.

Quantitative structure activity relationship model for predicting the depletion percentage of skin allergic chemical substances of glutathione

A quantitative model was developed to predict the depletion percentage of glutathione (DPG) compounds by gene expression programming (GEP). Each kind of compound was represented by several calculated structural descriptors involving constitutional, topological, geometrical, electrostatic and quantum-chemical features of compounds. The GEP method produced a nonlinear and five-descriptor quantitative model with a mean error and a correlation coefficient of 10.52 and 0.94 for the training set, 22.80 and 0.85 for the test set, respectively. It is shown that the GEP predicted results are in good agreement with experimental ones, better than those of the heuristic method.

Classification study of skin sensitizers based on support vector machine and linear discriminant analysis

The support vector machine (SVM), recently developed from machine learning community, was used to develop a nonlinear binary classification model of skin sensitization for a diverse set of 131 organic compounds. Six descriptors were selected by stepwise forward discriminant analysis (LDA) from a diverse set of molecular descriptors calculated from molecular structures alone. These six descriptors could reflect the mechanic relevance to skin sensitization and were used as inputs of the SVM model. The nonlinear model developed from SVM algorithm outperformed LDA, which indicated that SVM model was more reliable in the recognition of skin sensitizers. The proposed method is very useful for the classification of skin sensitizers, and can also be extended in other QSAR investigation.

Assessment of the relative skin sensitization potency of siloranes and bis-GMA using the local lymph node assay and QSAR predicted potency

Siloranes are silicon and oxirane (epoxy) containing monomers used for new dental composite development. The siloranes 3,4-epoxycyclohexylethyl-cyclopolymethylsiloxane (Tet-Sil) and bis-3,4-epoxycyclohexylethyl-phenyl-methylsilane (Ph-Sil) have in common cycloaliphatic epoxy moieties. The epoxy group is of concern in their biocompatibility since most epoxy compounds are known skin sensitizers. The objective of this study was to determine the in vivo skin sensitization potency of the siloranes in the local lymph node assay. A comparison was made with well-known chemical allergens, bis-GMA and DNCB. Female mice (CBA/CaJ) were exposed topically (dorsum of both ears) to several doses of acetone:olive oil in the ratio of 4:1 v/v. Doses were defined by a predictive structure-activity model (QSAR) for contact sensitization. Lymph node cell (LNC) proliferation was measured on the sixth day by incorporation of radioactive thymidine into DNA of lymph node cells. The effective concentration (EC3) that produced a 3-fold stimulation in LNC proliferation relative to controls was extrapolated from dose-response curves. DNCB was a strong sensitizer (EC3 = 0.06%). The EC3 values of Ph-Sil and bis-GMA were 19% and 45%, respectively, making these weak contact sensitizers. Tet-Sil did not increase lymph node proliferation when compared with controls. In contrast to Tet-Sil, the unpolymerized monomers Ph-Sil and bis-GMA have the capacity to induce LNC proliferation, characteristic of a T-cell mediated skin contact sensitization.