Relationship Between Adverse Outcome Pathways and Chemistry-Based In Silico Models to Predict Toxicity

Abstract

The current landscape of Adverse Outcome Pathways (AOPs) provides a means of organizing information relating to the adverse effects elicited following exposure to chemicals. As such, AOPs are an excellent driver for the development and application of in silico models for predictive toxicology allowing for the direct relationship between chemistry and adverse effects to be established. Information may be extracted from AOPs to support the creation of (quantitative) structure-activity relationships [(Q)SARs] and to increase confidence in grouping and read-across. Any part of an AOP can be linked to these various types of in silico models. There is, however, an emphasis on using information from known molecular initiating events (MIEs) to create models including 2-D and 3-D structural alerts, SARs and QSARs. MIEs can be classified according to the nature of the interaction, for example, covalent reactivity, oxidative stress, phototoxicity, chronic receptor mediated, acute enzyme inhibition, unspecific, physical and other effects. Different types of MIEs require different approaches to their in silico modeling. Modeling Key Events and Key Event Relationships is useful if they represent the rate limiting step or key determinant of toxicity. Modeling of metabolism and chemical interactions will become part of AOP networks, which are also driving species-specific extrapolation and respective adaptation of models. With more information and data being captured, in silico approaches will increasingly support the application of knowledge from AOPs to build weight of evidence and support risk assessment, for example, in the context of Integrated Approaches to Testing and Assessment (IATAs).

Introduction

As a gradual process, the toxicological testing of chemicals is undergoing a paradigm shift. For decades there has been a reliance on in vivo and latterly, to a more limited extent, in vitro testing of a chemical with the effects being recorded and used to allow for some type of risk assessment. The shift is toward an assessment of the (perturbation of) normal biological pathways relating to toxicity allowing for a mechanistic basis to understanding the effects of chemicals, which is often overlooked in the traditional testing paradigm. This has a number of advantages, not least in the replacement of in vivo testing with associated reduction in animal use and, in most cases, costs and the time involved. Advances in what is termed 21^st Century Toxicology are developing tools and techniques that provide information on pathways (and hence mechanisms) that are derived from cells and cell lines more relevant to humans and other target species, rather than relying on the extrapolation from surrogate test species.¹

The development of the (toxicological) pathway concept, not least stimulated by the 2007 National Research Council's report,² has seen a growth in the number of assays and hence data available, for example, those directly associated with pathways such as from high throughput/content assays such as Tox21/ToxCast³ and omics^4–6 or those that may be applied indirectly from existing in vivo and other data.^7–9 These assays have provided an explosion in the number of pathway related data available, which could form the basis of either strategies to predict toxicity, or be used to develop computational models. However, there has been no overarching means of interpreting, rationalizing, or utilizing these data. As such, the Adverse Outcome Pathway (AOP) concept was developed in part as a response to the call for toxicology to become pathway orientated, but also to assist in the utilization of data from such assays.¹⁰

It is appreciated that AOPs are a conceptual framework to support the prediction of toxicity, a full description of AOPs is not given here but is available from various authors.^10–19 In practical terms AOPs will assist in the organization of data, assays, mechanistic knowledge, and models to predict toxicity within the paradigm of 21^st Century Toxicology. The ability to use the pathway-derived data to extract further information and knowledge is one of their advantages, especially when they can be formalized into computational models.^20,21 When associated with chemical structure, these models, also called in silico approaches, can provide a direct linkage between chemistry and adverse effect, leveraging the content of the AOP to support the meaning and interpretation of the model.^22,23

In silico models to predict toxicity vary from 2-D structural alerts derived from structure-activity relationships (SARs), to more complex quantitative structure-activity relationships (QSARs) capable of predicting toxic potency.²⁴ This article explores the linkage of these models, as well as grouping and read-across, to AOPs. At the outset it is already well acknowledged that AOPs can support computational modeling.²¹ In addition, the use of AOPs to support computational modeling deriving structural alerts^25–28 for toxicity prediction, or as part of a grouping strategy leading to read-across and for QSAR development, is well established.²⁹ Other computational approaches, beyond the ab initio risk assessment consideration, that utilize and extend the AOP framework are the development of Integrated Approaches to Testing and Assessment (IATAs).^{13,18,30–38}

The Links Between In Silico Models and the Different Steps of an AOP

Figure 1 illustrates that models, or in silico approaches, may potentially be utilized at all stages of the AOP to provide knowledge and information (it should be noted that the structure of the generic AOP shown in Figure 1 is illustrative only and many AOPs do not include exposure or proceed to the ecosystem level). There are many purposes to the use of models within the AOP framework; these include:

• providing data and information,

• capturing the knowledge of the AOP, and

• extrapolation and prediction.

FIG. 1.

Schematic of a generic Adverse Outcome Pathway (from source to ecosystem) with the types of in silico models that may be associated with each Key Event. AOP, Adverse Outcome Pathway; (Q)SARs, (quantitative) structure-activity relationships; PBPK, physiologically based pharmacokinetic; QIVIVE, quantitative in vitro to in vivo extrapolation.

Computational, or in silico, models for toxicity prediction are many and varied. The focus of this article is the relationship of chemistry-based models for toxicity prediction with AOPs. Briefly, as discussed in more detail below, in silico models are especially useful to capture and predict the molecular initiating event (MIE) of an AOP, such as binding to biological molecules or receptors, and thus predict which chemicals can potentially be associated with adverse effects triggered by the MIE. (Q)SARs can also model Key Events or Key Event Relationships further downstream in an AOP, which is relevant for toxicity prediction in particular if these events are rate-determining steps in an AOP. In this case biokinetic modeling will support the prediction of which steps in an AOP actually take place in specific exposure situations, and can contribute to the prediction of which target organs a chemical will be distributed to and the actual target organ concentrations. Modeling exposure related to the source and models determining, for example, oral or skin absorption or bioavailability as a whole, will support quantification of the effects occurring as a result of the AOP. Overall, a higher level of modeling, such as multiscale modeling, will be required to capture all elements of the AOP to link exposure to risk. Extending the consideration of adverse outcomes to the individual, population, or ecosystem level will allow for models to be created for more complex adverse effects.

Figure 2 summarizes the main differences where models may drive the development of AOPs and vice versa. It illustrates that on one hand AOPs can support the building of predictive models, to base them on mechanistic information and give them relevance for toxicity assessment, and on the other hand, computational models can help to compile AOP pathways. When both robust mechanistic information and computational models are available, this will allow for the better prediction of relevant adverse effects or their use in an IATA. As such, the ultimate aim of building models and organizing mechanistic information within the AOP framework is to provide relevant prediction of adverse effects to allow for adequate chemical safety assessment.

FIG. 2.

Main differences where models may drive the development of AOPs and vice versa.

It is important to also consider that there will be a cyclical process whereby models will be developed on the basis of a (putative) AOP, which will then be refined as more information is passed back to the model—likewise this will assist in the refinement of the AOP. More broadly, the relationship between AOPs and in silico chemistry based models can be considered as being:

(i) models using information derived from the AOP,

(ii) models that support the use or application of information from the AOP, for example an AOP being used to provide evidence for the ab initio risk assessment of a substance and build a mode-of-action hypothesis,³⁹

(iii) AOPs supporting the mechanistic interpretation of a model or providing evidence to build a weight of evidence to support risk assessment (cf. read-across case studies).^40–43

This article attempts to provide an overview of the types of models that may be applied at different stages of the AOP, taken in the broadest context from sources to ecosystem, with a particular focus on the mechanistic and toxicological aspects. However, this article focuses on the mechanistic aspects of the AOP, it does not consider exposure modeling, in silico models for the apical endpoint [e.g., traditional (Q)SARs] or ecosystem level models.

Predictive Models Based on AOP-Related Mechanistic Information

Models of the MIE

Information from the MIE is one of the key drivers for in silico models derived from an AOP.^25–28 Before computational modeling approaches are considered, it should be recognized that there are different types of MIEs and each may require a different modeling approach. These have been summarized in general terms in Table 1, along with examples and the types of models that can be derived from the different types of MIE.^{22,23,44–74} Table 1 provides by no means an exhaustive list and practitioners will no doubt wish to expand this according to their needs and experience. The key to understand why this is important is in the concept of, if the MIE is known, then an appropriate model can be developed for it.

Table 1.

Types of Molecular Initiating Events with Associated Adverse Outcomes and Types and Examples of Appropriate Modeling Approaches

Type of MIE	Description	Adverse outcome	Appropriate type(s) of modeling approach(es)	Examples of models
Covalent reactivity	Covalent binding of the compound with a biological macromolecule, for example DNA, cellular membrane proteins, immunoproteins	Mutagenicity, skin sensitization, respiratory sensitization, liver fibrosis, acute toxicity evaluated above baseline	2-D structural alerts for specific molecular fragments, quantum chemical calculations of reactivity	DNA binding⁴⁴; protein binding⁴⁵; respiratory sensitization⁴⁶; DFT calculations^47–49
Free radical/oxidative stress	Cellular or tissue damage caused by free radicals promoted by, for example, redox cycling	Mutagenicity, tissue damage, aging, mitochondrial toxicity	2-D structural alerts for specific molecular fragments, calculation of redox potential	2-D alerts for excess aquatic toxicity^50–52; mitochondrial toxicity⁵³
Phototoxicity	Damage caused by a reactive molecule or radical following excitation by UV light	Mutagenicity, tissue damage, aging	2-D structural alerts for specific molecular fragments, quantum chemical calculations of chemical stability	2-D alerts and QSARs for excess aquatic (photo)toxicity^54–56
Chronic receptor mediated	Stimulation or disruption of a normal physiological process, for example, hormonal control, through (antagonistic or agonistic) binding (usually reversible) to a receptor	Many and varied adverse outcomes, for example, disruption of endocrine function leading to reproductive impairment, cancer etc.	2-D alerts for scaffolds associated with binding. 3-D toxicophores, receptor-ligand docking, molecular dynamics, and other associated molecular modeling techniques	2-D alerts for nuclear receptor binding leading to steatosis^22,23; 3-D toxicophores for PPARγ binding leading to steatosis^57,58; 2-D alerts and QSARs leading to reproductive toxicity and other endocrine effects^59–61
Acute enzyme inhibition	Irreversible inhibition of a physiological enzyme, for example, acetylcholinesterase	Rapid acute toxicity elevated above baseline	2-D structural alerts for specific molecular fragments. 3-D toxicophores	2-D and 3-D QSARs for excess aquatic toxicity as a result of enzyme inhibition^62–64
Unspecific	Unspecific events that occur at any relevant site, for example, reversible membrane disruption	Narcosis referring to basal cytotoxicity, resulting in anesthesia and lethality, phospholipidosis leading to liver failure	Relevant physicochemical properties, for example, log P in aquatic environment, vapor pressure in air. 2-D structural alerts for specific molecular fragments	QSARs for nonpolar narcosis to aquatic species^65–67; 2-D alerts for phosopholipidosis^68,69
Physical	Disruption of membranes due to the (physical/chemical) characteristics of a molecule	Skin, eye, nasal, respiratory (other membrane) irritation and corrosion	2-D structural alerts for specific molecular fragments. Measures of acidity/basicity; surfactant activity, for example, critical micelle concentration	2-D alerts for skin/eye irritation/corrosion^70,71; pH for irritation/corrosion⁷²; QSAR for irritation⁷³
Other^a	Any other effect that may be related to promoting an adverse outcome, for example, weather, temperature or anthropogenic factors	Any, however usually associated with population decline	Any relevant effect or property	Climate as a stressor⁷⁴

It is acknowledged that AOPs in this category do not have a chemical interaction. This category of MIE is included in this table to enable any framework to categorize AOPs to be fully inclusive of all possible interactions. It should be noted that non-chemical interactions may need to be defined separately.

DFT, Density Functional Theory; MIE, molecular initiating event; QSAR, quantitative structure-activity relationship; PPAR, peroxisome proliferator-activated receptor.

On the basis of knowledge of different types of MIEs, it is thus possible to classify AOPs with regard to the type of MIE. To illustrate the types of MIEs, Table 2 lists the citable AOPs (at the time of preparation of the article) listed on the AOP Wiki (https://aopwiki.org),⁷⁵ which is one component of the OECD-sponsored AOP Knowledge Base (AOP-KB, http://aopkb.org).⁷⁶ This wiki represents the central repository for all AOPs developed as part of the OECD AOP development effort by the Extended Advisory Group on Molecular Screening and Toxicogenomics. The AOP Wiki is not, of course, a complete list of AOPs with many others becoming available. However, what is clear from Table 2 is that the majority of citable AOPs on the AOP Wiki are for what is termed “chronic receptor mediated” toxicity, with a smaller number classified as being “covalent reactivity.” This is probably due to the AOPs being defined, at least at the start of the process of their compilation, around common, well-known and familiar modes of action. Table 2 was also supplemented by literature examples of AOPs where none were available as being citable from the AOP Wiki. It must be appreciated that these literature AOPs are likely to be less developed (in the formal sense of the AOP Wiki) at the time of writing.^{12,29,62,67,77,78}

Table 2.

Examples of Currently Available Adverse Outcome Pathways Classified According to the Type of Molecular Initiating Event

AOP type and title	MIE of AOP	Adverse outcome	OECD project number or reference; AOP Wiki ID
Covalent reactivity
Covalent protein binding leading to skin sensitization	Covalent binding to immunoprotein	Skin sensitization	OECD 1.1 ID: 40
Alkylation of DNA in male premeiotic germ cells leading to heritable mutations	DNA alkylation	Heritable mutations	OECD 1.11 ID: 15
Protein alkylation leading to liver fibrosis	Protein alkylation	Liver fibrosis	OECD 1.14 ID: 38
AFB1: mutagenic mode-of-action leading to hepatocellular carcinoma	Formation of pro-mutagenic DNA Adducts	Hepatocellular carcinoma	OECD 1.8 ID: 46
Free radical/oxidative stress
No AOP currently citable on the AOP Wiki (although some are in draft form) or available in the literature
Phototoxicity
No AOP currently citable on the AOP Wiki or available in the literature
Chronic receptor mediated
Inhibition of thyroperoxidase and subsequent adverse neurodevelopmental outcomes in mammals	Inhibition of thyroperoxidase	Neurodevelopmental outcomes	OECD 1.10ID: 42
Disruption of VEGFR signaling leading to developmental defects	Inhibition of VegfR2	Developmental outcomes	OECD 1.6ID: 43
Sustained AhR activation leading to rodent liver tumors	Binding to AhR	Liver tumors	OECD 1.7ID: 41
Androgen receptor agonism leading to reproductive dysfunction	Agonism of androgen receptor	Reproductive dysfunction	OECD 1.12ID: 23
Aromatase inhibition leading to reproductive dysfunction	Aromatase inhibition	Reproductive dysfunction	OECD 1.12ID: 25
Estrogen receptor antagonism leading to reproductive dysfunction	Binding to the estrogen receptor	Reproductive dysfunction	OECD 1.12ID: 30
Chronic binding of antagonist to NMDARs during brain development leads to neurodegeneration with impairment in learning and memory in aging	Binding to NMDAR	Neuroinflammation leading to neurodegeneration	OECD 1.13ID: 12
Chronic binding of antagonist to NMDARs during brain development induces impairment of learning and memory abilities	Binding to NMDAR	Impairment of learning and memory abilities	OECD 1.22ID: 13
Aromatase (Cyp19a1) reduction leading to impaired fertility in adult female	Binding to aromatase	Impaired fertility	OECD 1.21ID: 7
PPARα activation in utero leading to impaired fertility in males	Binding to PPAR	Impaired fertility	OECD 1.21ID: 18
Binding of agonists to ionotropic glutamate receptors in adult brain causes excitotoxicity that mediates neuronal cell death, contributing to learning and memory impairment	Binding to glutamate receptor	Learning and memory impairment	OECD 1.23ID: 48
Inhibition of the mitochondrial complex I of nigra-striatal neurons leads to parkinsonian motor deficits	Binding of inhibitor to NADH-ubiquinone oxidoreductase (complex I),	Motor function, impaired	OECD 1.33ID: 3
Acute enzyme inhibition
Inhibition of acetylcholinesterase leading to lethality	Irreversible binding to acetylcholinesterase	Lethality	^12,62
Unspecific
Reversible membrane disruption leading to anesthesia, for example, nonpolar narcosis, basal cytotoxicity	Unspecific membrane disruption	Anesthesia, lethality	^29,67,77,78
Physical
Intracellular acidification induced olfactory epithelial injury leading to site of contact nasal tumors	Decrease in intracellular pH	Contact nasal tumors	OECD 2.7
Other
No AOP currently citable on the AOP Wiki (although some are in draft form) or available in the literature

Where available “citable” AOPs from the AOP Wiki (https://aopwiki.org) have been used in preference to those available in the literature.

It is striking from Table 1 that the majority of models derived from AOPs utilize 2-D approaches to capture molecular fragments responsible for toxicity. There is a rich history of such “structural alerts” and their conversion into usable computational models.²⁰ There are many good reasons for this, for example, they are easy to define and comprehend—hence aiding in their transparency. Technologies such as SMARTS strings have made chemistry fragments easy to code and be handled computationally.⁷⁹ They have also been developed into a number of commercially and freely available expert systems. Most notable among the freely available systems is the OECD QSAR Toolbox, which for some endpoints,⁸⁰ for example skin sensitization, provides a direct linkage to the AOP and the possibility to build a read-across argument on the basis of data from the AOP.^81,82 They are particularly suited when a structural fragment drives a particular chemical and/or biological interaction, for example, covalent binding to DNA or proteins.^44,45 The disadvantage is that 2-D structural alerts implicitly are not able to capture 3-D properties and configurations that are required for receptor binding. 2-D alerts, for example, those coded as SMARTS strings, can be extended and made more sophisticated when considered as “chemotypes”, i.e., alerts extended with further information on molecular structure such as atomic charges; these chemotypes may be coded in the so-called Chemical Structure Reactivity Markup Language (CSRML).^3,83

Receptor mediated MIEs leading to toxicity are more difficult to capture in terms of models. However, there is a growing appreciation of the needs for such models to assist in the prediction of chronic toxicity. 3-D or conformationally dependent properties can, to a limited extent, be captured with extended SMARTs strings that represent particular scaffolds.^22,23 However, this is a relatively crude approach that may be suited for grouping and read-across but less for toxicity prediction when understanding stereoisomerism may be important. Chemotypes are more suitable for the capture and modeling of 3-D effects as they can supplement 2-D structural alerts with other properties.⁸³ Most suited, although requiring the greatest level of expertise for application, are techniques derived from drug design, for example, molecular modeling of the receptor-ligand interaction and/or development of toxicophores.^57,58 There are also many reported studies (beyond the scope of this article) on modeling effects associated with endocrine disruption—most notably binding to the estrogen receptor. As an example, the recent CERAPP project, where a large variety of mostly QSAR-type models were developed, demonstrates the relevance of this approach.⁸⁴ Extending the endocrine disruption paradigm, the work of Wu et al.⁵⁹ mapped and compiled known mechanisms of developmental toxicity, supporting them with data and structural alerts. This is a scheme that allows for the interpretation and prediction of developmental toxicity rapidly, but at this time is, at best, only partially supported by AOPs. However, this approach maps out where development of AOPs is required.

There are a variety of modeling approaches relating to toxicities associated with unspecific MIEs, that is where there is no single definable interaction analogous to receptor binding or covalent interactions. For instance, models may be based on 2-D properties associated with giving the molecule the capability to act in the required manner, for example, the amine functionality promoting membrane accumulation leading to phospholipidosis.^68,69 Alternatively, models may be related to the properties that govern an unspecific effect, for example hydrophobicity being the determining feature for nonpolar narcosis.⁶⁶

Overall, there is a need to identify the MIE and appropriate modeling techniques for it—this is crucial information that can be derived from an AOP. (Q)SARs should be benchmarked against the MIE to ensure that the modeling approach is valid for the endpoint of interest. Inappropriate modeling approaches run the risk of spurious and overfit models.

Models of key events and key event relationships

As indicated in Figure 1, the modeling of Key Events is further along the AOP toward apical endpoints and may be considered more relevant and quantitative than the MIE. Models may be developed for data derived from assays relating to individual Key Events, these may, for instance, be in the form of QSARs. However, there are two distinct issues in the modeling of data for Key Events, namely the difficulty in obtaining such data (although initiatives such as ToxCast and other High Throughput Screening projects may assist here)³ and their relevance, especially when considered in isolation from the MIE, in comparison to, for instance, the MIE itself. However, when relevant data are available that are founded in the preceding Key Events within the AOP, the modeling of Key Events and Key Event Relationships may provide a more quantitative prediction of (adverse outcome) potency providing they are the rate limiting step. In addition, should a Key Event or Key Event Relationship not occur routinely following interaction with the MIE, and require further stimulation, for example, another MIE or reaching a critical concentration (a Point of Departure for the Key Event), then models may provide more useful information than one based on the MIE alone. However, should the Key Event provide no more meaningful information than the MIE, then it may be more appropriate to model the MIE as it is more fundamental and likely to be more rapid and less expensive to obtain data for.

Any models for Key Events will be similar to those for MIEs, that is, 2-D structural alerts for specific molecular fragments, toxicophores for receptor interactions. However, there are relatively few assay data available relevant to Key Events and even fewer, if any, models. One good source of information has been from the Ames test and related assays for genotoxic carcinogenicity. The Ames test in this instance is a good surrogate for the MIE involving interaction with DNA but less likely to be appropriate for the later Key Events in this AOP. Another source of data would be for skin sensitization where there is a well developed AOP¹³ and where the AOP has been converted into an IATA.³³ Despite the proliferation of assays for the different stages of the AOP, with the exception of models for protein reactivity,^49,85 no viable models for example for h-CLAT or other intermediate effect events have been published. Thus, in silico models would have the potential to eventually replace assays within the IATA, but have not yet been fully recognized, such models would assist in hazard identification and ultimately risk assessment. The reliability and relevance of in silico models to predict an endpoint within an IATA will be increased if the models are mechanistically based and cover different steps of an acknowledged AOP in the context of a consensus approach, that is, concordant results with independent prediction models.

There is also the possibility to model Key Event Relationships, such models may assist in the quantification and understanding of the relationship. The benefit here is that should the Key Event Relationship be the rate limiting step then a model for that specific relationship, which may resemble a QSAR or quantitative activity-activity relationship (QAAR), could hold the key to making a quantitative prediction of potency. Currently there are few, or no, examples of (Q)SAR or QAAR models for Key Event Relationships, although some in silico models for Key Event Relationships are becoming available, especially in the form of quantitative AOPs (qAOPs)—as described in the next section.

Ongoing efforts to formalize relationships in AOPs through ontologies⁸⁶ and to capture response-response relationships and corresponding data, for example, in the Effectopedia module of the AOP-KB (http://effectopedia.org)⁸⁷ will further support future model development in this area.

Models related to the whole AOP

It is conceivable that the ultimate aim of modeling the AOP may result in the development of an algorithm, or sequence of models,²¹ to include all steps of an AOP. However, it is recognized that this is a long-term aspiration and will not, necessarily, be required if there is the possibility to predict the adverse outcome itself with a defined level of confidence. Currently, there are no chemistry-based approaches, i.e. (Q)SARs, that attempt to describe the AOP as whole. In the longer term, there may be value in this as it may assist in the implementation of models from the AOP in the form of IATA.^88,89 However, there will be limited value going beyond chemistry-based models for the MIE and possibly the rate-limiting Key Event Relationship. There is increasing interest in qAOPs that attempt to formalize the Key Event Relationships from the MIE to the Adverse Outcome, for example, Conolly et al.⁸⁹ developed a qAOP that linked the inhibition of cytochrome P450 19A aromatase (the MIE) to population-level decreases in the fathead minnow. The qAOP itself consisted of three linked computational models for relevant parts of the AOP.

A further approach that has yet to be fully applied to AOPs is that of multilevel or multiscale modeling.⁹⁰ These approaches attempt to derive interconnected models for each component of the AOP in a manner that is representative of the overall AOP. As such it is a data hungry approach. These models may be fed with data from public data sources such as ChEMBL^{22,23,91–93} and ToxCast/Tox21.⁹⁴ The use of ToxCast data in this context remains an opportunity and a challenge, with more data and a better appreciation of how to use the data required.⁹⁵

Specific Considerations for In Silico Modeling with Regard to AOPs

Transformations and interaction of chemicals

Until now, metabolic activation has seldom been considered as part of an AOP, however, it is necessary for chemistry-based in silico modeling (and the use of information from an in vitro assay) if a realistic toxicological profile is required and only the parent structure is available. For instance, it is important to determine whether the parent molecule will be metabolized into a more toxic form or detoxified. A large number of in silico approaches are available to predict potential metabolites,⁹⁶ some freely available and others may require payment. The performance of such predictive methods is not always ideal with over-prediction of metabolites being one problem, and the difficulty to identify stable metabolites and kinetics/rates of transformation being another.⁹⁷ There is, however, the possibility that knowledge of metabolic events in AOPs may provide better knowledge to support improved toxicity prediction.

Another possibility to use knowledge from AOPs to support the capture of information and development of in silico models is with effects such as chemical interactions. In this case, should a key stage of an AOP require the inhibition of an enzyme (either from the target or another molecule), this knowledge could be incorporated into a model. As yet, there are no approaches that have used the information from an AOP in this way—this is something that would be of immense benefit for understanding and modeling drug–drug interactions as well as predicting effects from combined exposure to chemicals, taking into consideration possible synergistic effects—even over time—which are not accounted for when evaluating chemicals individually.

As AOPs evolve naturally from linear constructs to networks of inter-related effects,²¹ metabolism and interactions will become integral to the development of AOP networks within an ontology framework, a formalized way to organize AOP knowledge and capture AOP relationships.⁸⁶ This provides the possibility, for example, to include models for metabolism, particularly the ability to bind to a CYP or enzyme and the rate of binding/inhibition. In addition, models will be required to predict relevant metabolites that may go elsewhere in the AOP network, or to completely independent AOPs. The other possibility is that knowledge of metabolism, metabolite formation, and rates of formation that is contained within the AOP network will provide information for the models themselves, thus the AOP will drive the model.

Interspecies relationships of toxicity and sensitivity

AOPs provide a rational means to extrapolate toxicity across species. The relative similarity of acute aquatic potency within groups of compounds acting by a similar mechanism of action, especially nonpolar narcosis, has been known for many years⁹⁸ and can be applied to successful QSAR development.²⁹ Similarly, the need to take care of other mechanisms of action, particular those associated with species-specific metabolism⁹⁹ or reactivity¹⁰⁰ is important and can be related to the individual AOPs. While generalist extrapolation approaches may work for lethal potency, more sophisticated modeling will be required for receptor mediated responses. Knowledge of the essential receptors in an AOP can help drive the extrapolation of AOPs from one species to another. Thus, while AOPs are chemical-agnostic they are also species-specific and therefore can drive the adaption/applicability of predictive models. The web-based SeqAPASS tool is one such approach, utilizing information on receptor homology to allow for extrapolation of effects, being wholly supported by the AOP.^101,102

Models for exposure and bioavailability—adding value to AOPs

While not strictly part of the AOP, computational models for exposure and bioavailability are useful for their implementation.¹⁰³ These can impact at various points while using information from an AOP. Models for exposure and bioavailability also vary in their complexity. Another area of progress is the development of Aggregate Exposure Pathways that will provide information to derive models from.¹⁰⁴

There are a number of SAR approaches that can be used to screen out compounds that are not likely to reach the site of the MIE. Very commonly used (especially as a drug design tool) is the Lipinski Rule of 5¹⁰⁵ and various adaptions.¹⁰⁶ The premise here is that a small number of easily calculated properties can identify compounds with low solubility which is taken as being indicative of poor oral absorption—thus, if this is required for systemic exposure, then risk assessment may be able to discount oral exposure as a likely route to stimulate the AOP. In a similar manner, Ates et al. developed a small number of physicochemical properties that are seen to be determinants of poor absorption through the skin.¹⁰⁷

There are also QSARs for effects such as bioavailability that may be used as part of the risk assessment process. In addition, QSARs can provide an estimate of exposure following absorption through the skin,^107,108 blood-brain barrier,¹⁰⁹ cornea,¹¹⁰ and various other membranes.⁹⁶ While such QSARs may be useful, to assess organ level concentrations following different types of exposure and doses requires the use of physiologically based kinetic models.^111,112

AOP-Derived Models Driving Grouping and Domain Definition

In addition to supporting models such as (Q)SARS, the information from MIEs is a key means of grouping compounds providing a mechanistic basis and transparency.^113,114 In silico 2-D profilers of structural alerts, built on knowledge of MIEs, have provided a means to group compounds for a number of adverse outcomes including skin sensitization⁷⁹; respiratory sensitization⁴⁶; phospholipidosis^68,69; mutagenicity^44,115; hepatotoxicity¹¹⁶; reproductive toxicity,⁵⁹ and testicular toxicity.¹¹⁷ Successful grouping allows for read-across to fill data gaps for these types of endpoints.

The MIE can also be used to define the chemical domain of an AOP through intelligent testing using assays derived around the MIE. For instance, the definition of mechanistically derived domains of reactivity, as measured by in chemico testing, has been shown to assist in the identification of chemical domains associated with protein reactivity, which in turn may be related to effects such as skin sensitization or elevated acute toxic potency.^118,119

The value of AOPs to support read-across was demonstrated in a series of case studies.^40–43 The case studies demonstrated how the read-across hypothesis and justification were strengthened with knowledge of data from New Approach Methodologies that were related to an AOP.¹²⁰

Conclusions and Recommendations

AOPs provide a framework to organize information within the context of 21^st Century Toxicology. As such, they are a good driver for the creation, development, and application of in silico models to predict toxicity. Indeed, AOPs provide the direct mechanistic relevance and transparency that is a prerequisite for in silico models of toxicity. The ability to be able to reference back to an AOP has been shown to be crucial for the justification of grouping and read-across and to allow for the development of IATA. As further information and data are captured, in silico models provide the means to create knowledge and apply it in a rational manner to predict toxicity.

Currently, most in silico models from AOPs are derived from the MIE. Qualitative models can be derived from knowledge of the MIE, in theory with limited knowledge and without data to support the hypothesis, for example, an alert for skin sensitization may be developed on the basis of only a single data point if there is a clear mechanistic (i.e., organic chemistry) understanding. Quantitative models for the MIE, as well as for Key Events and Key Event Relationships require a more complete data set with information from a greater number of compounds covering a range of activity and properties. MIEs, in terms of modeling, can be classified into a number of types (Tables 1 and 2). It is useful to classify MIEs in this manner. Capturing the type of MIE will allow for possibilities for modeling from an AOP to be identified; these could be stored in the AOP Wiki in the first instance and then progressed to Effectopedia and tools such as the OECD QSAR Toolbox. To maximize the use of information from AOPs there should be a shift to storing information on the MIE in a manner that could be translated into an in silico model, for example, classifying MIEs in tools such as Effectopedia and providing a direct linkage from there to a chemistry-based model. Currently, most in silico models derived from the MIE are based on 2-D descriptions of chemistry (with some notable exceptions). Most models are for MIEs associated with receptor binding, with covalent interactions also well represented.

Key Events and Key Event Relationships may also prove useful for modeling, however, with notable exceptions such as receptor binding data and endpoint data such as those for the Ames assay, there is a lack of data. The lack of data is particularly acute for “intermediate” Key Events and Key Event Relationships, that is those between the MIE and the adverse outcome. With increasing data becoming available in the framework of 21^st Century Toxicology and high-throughput measurements, such data gaps may be filled in the future. However, they will only be useful if they represent the rate limiting step or key determinant of toxicity. These should form the basis of QSARs in particular as there is a desire to make the predictions more quantitative. The accurate prediction from chemical structure of metabolites and their rates of formation have historically been very difficult; this is an area where improvements are required, especially as networked AOPs become a reality.

Modeling of the whole AOPs, or qAOPs, will also require further data. At the moment a multitude of full chemistry-based computational models for a complete AOP seems unlikely. However, consideration of the AOP framework, especially as the new and/or updated tools become available, will both support model development and allow for the identification of gaps where either models or AOPs are required.

In conclusion, ideally in silico models will ultimately allow for the complete replacement of toxicological testing whereby only a chemical structure is required to make a hazard and/or risk assessment—however, we are a long way from achieving this. As the frameworks provided by AOPs (or whatever framework is utilized) to organize information and data become more sophisticated, the needs of modeling and the intimate and intricate relationship between the AOP and the models should not be forgotten but should be at the heart of AOP development and data capture.

Footnotes

Author Disclosure Statement

No competing financial interests exist.

References

National Academies of Sciences, Engineering, and Medicine. Using 21st Century Science to Improve Risk-Related Evaluations. Washington, DC: The National Academies Press; 2017.

National Research Council.. Toxicity Testing in the Twenty-first Century: A Vision and a Strategy. Washington, DC: The National Academies Press; 2007.

Richard

, Judson

, Houck

, et al. ToxCast chemical landscape: Paving the road to 21st Century toxicology. Chem Res Toxicol, 2016:29; 1225–1251.

Ganter

, Snyder

, Halbert

, et al. Toxicogenomics in drug discovery and development: Mechanistic analysis of compound/class-dependent effects usign the DrugMatrix^® database. Pharmacogenomics, 2006:7; 1025–1044.

van Ravenzwaay

, Sperber

, Lemke

, et al. Metabolomics as read-across tool: A case study with phenoxy herbicides. Reg Toxicol Pharmacol, 2016:81; 288–304.

van Ravenzwaay

, Herold

, Kamp

, et al. Metabolomics: A tool for early detection of toxicological effects and an opportunity for biology based grouping of chemicals-From QSAR to QBAR. Mutat Res, 2012:746; 144–150.

Gaulton

, Bellis

, Bento

, et al. ChEMBL: A large-scale bioactivity database for drug discovery. Nucleic Acids Res, 2012:40; D1100–D1107.

Bento

, Gaulton

, Hersey

, et al. The ChEMBL bioactivity database: An update. Nucleic Acids Res, 2014:42; D1083–D1090.

Wang

, Xiao

, Suzek

, et al. PubChem: A public information system for analyzing bioactivities of small molecules. Nucleic Acids Res, 2009:37; W623–W633.

10.

Ankley

, Bennett

, Erickson

, et al. Adverse Outcome Pathways: A conceptual framework to support ecotoxicology research and risk assessment. Environ Toxicol Chem, 2010:29; 730–741.

11.

Volz

, Belanger

, Embry

, et al. Adverse Outcome Pathways during early fish development: A conceptual framework for identification of chemical screening and prioritization strategies. Toxicol Sci, 2011:123; 349–358.

12.

Schultz

. Adverse Outcome Pathways: A way of linking chemical structure to in vivo. toxicological hazards. In: In Silico Toxicology: Principles and Applications, . MTD Cronin, JC Madden (eds); pp. 346–371. Cambridge: Royal Society of Chemistry; 2010.

13.

Delrue

, Sachana

, Sakuratani

, et al. The Adverse Outcome Pathway concept: A basis for developing regulatory decision-making tools. Altern Lab Anim, 2016:44; 417–429.

14.

Burden

, Sewell

, Andersen

, et al. Adverse Outcome Pathways can drive non-animal approaches for safety assessment. J Appl Toxicol, 2015:35; 971–975.

15.

Vinken

. Adverse Outcome Pathways and drug-induced liver injury testing. Chem Res Toxicol, 2015:28; 1391–1397.

16.

LaLone

, Ankley

, Belanger

, et al. Advancing the Adverse Outcome Pathway framework—An international horizon scanning approach. Environ Toxicol Chem, 2017:36; 1411–1421.

17.

Benigni

, Battistelli

, Bossa

, et al. Endocrine disruptors: Data-based survey of in vivo tests, predictive models and the Adverse Outcome Pathway. Reg Toxicol Pharmacol, 2017:86; 18–24.

18.

OECD.. Guidance Document for the Use of Adverse Outcome Pathways in Developing Integrated Approaches to Testing and Assessment (IATA). Series on Testing & Assessment No. 260; ENV/JM/MONO(2016)67. Paris: OECD; 2016.

19.

Vinken

, Knapen

, Vergauwen

, et al. Adverse Outcome Pathways: A concise introduction for toxicologist. Arch Toxicol, 2017 [Epub ahead of print]; DOI: https://doi.org/10.1007/s00204-017-2020-z.

20.

Cronin

MTD

, Enoch

, Mellor

, et al. In silico prediction of organ level toxicity: Linking chemistry to adverse effects. Toxicol Res, 2017:33; 173–182.

21.

Wittwehr

, Aladjov

, Ankley

, et al. How Adverse Outcome Pathways can aid the development and use of computational prediction models for regulatory toxicology. Toxicol Sci, 2017:155; 326–336.

22.

Mellor

, Steinmetz

, Cronin

MTD

. Using molecular initiating events to develop a structural alert based screening workflow for nuclear receptor ligands associated with hepatic steatosis. Chem Res Toxicol, 2016:29; 203–212.

23.

Steinmetz

, Mellor

, Meinl

, et al. Screening chemicals for receptor-mediated toxicological and pharmacological endpoints: Using public data to build screening tools within a KNIME workflow. Mol Inform, 2015:34; 171–178.

24.

Cronin

MTD

, Madden

(eds) In Silico Toxicology: Principles and Applications. Cambridge: The Royal Society of Chemistry; 2010.

25.

Allen

THE

, Goodman

, Gutsell

, et al. A history of the molecular initiating event. Chem Res Toxicol, 2016:29; 2060–2070.

26.

Sanderson

, Simpson

, Cubberley

, et al. Mechanistic understanding of molecular initiating events (MIEs) using NMR spectroscopy. Toxicol Res, 2016:5; 34–44.

27.

Allen

THE

, Goodman

, Gutsell

, et al. Defining molecular initiating events in the Adverse Outcome Pathway framework for risk assessment. Chem Res Toxicol, 2014:27; 2100–2112.

28.

Gutsell

, Russell

. The role of chemistry in developing understanding of adverse outcome pathways and their application in risk assessment. Toxicol Res, 2013:2; 299–307.

29.

Ellison

, Piechota

, Madden

, et al. Adverse Outcome Pathway (AOP) informed modeling of aquatic toxicology: QSARs, Read-Across, and interspecies verification of modes of action. Environ Sci Technol, 2016:50; 3995–4007.

30.

Ezendam

, Braakhuis

, Vandebriel

. State of the art in non-animal approaches for skin sensitization testing: From individual test methods towards testing strategies. Arch Toxicol, 2016:90; 2861–2883.

31.

Basketter

. Skin sensitisation, Adverse Outcome Pathways and alternatives. Altern Lab Anim, 2016:44; 431–436.

32.

Schultz

, Dimitrova

, Dimitrov

, et al. The Adverse Outcome Pathway for skin sensitisation: Moving closer to replacing animal testing. Altern Lab Anim, 2016:44; 453–460.

33.

OECD.. Guidance Document on the Reporting of Defined Approaches and Individual Information Sources to Be Used Within Integrated Approaches to Testing and Assessment (IATA) for Skin Sensitization. Series on Testing & Assessment No. 256 ENV/JM/MONO(2016)29. Paris: OECD; 2016.

34.

Fitzpatrick

, Patlewicz

. Application of IATA—A case study in evaluating the global and local performance of a Bayesian network model for skin sensitization. SAR QSAR Environ Res, 2017:28; 297–310.

35.

Worth

, Patlewicz

. Integrated approaches to testing and assessment. In: Validation of Alternative Methods for Toxicity Testing. Eskes

, Whelan

(eds); pp. 317–342. Switzerland: Springer; 2016.

36.

Edwards

, Tan

, Villeneuve

, et al. Adverse Outcome Pathways—Organizing toxicological information to improve decision making. J Pharmacol Exp Ther, 2016:356; 170–181.

37.

Tollefsen

, Scholz

, Cronin

, et al. Applying Adverse Outcome Pathways (AOPs) to support Integrated Approaches to Testing and Assessment (IATA). Reg Toxicol Pharmacol, 2014:70; 629–640.

38.

Perkins

, Antczak

, Burgoon

, et al. Adverse outcome pathways for regulatory applications: Examination of four case studies with different degrees of completeness and scientific confidence. Toxicol Sci, 2015:148; 14–25.

39.

Berggren

, White

, Ouedraogo

, et al. Ab initio chemical safety assessment: A workflow based on exposure considerations and non-animal methods. Comput Toxicol, 2017 (In Press). DOI: https://doi.org/10.1016/j.comtox.2017.10.001.

40.

Przybylak

, Schultz

, Richarz, et al. Read-across of 90-day rat oral repeated-dose toxicity: A case study for selected β-olefinic alcohols. Comput Toxicol, 2017:1; 22–32.

41.

Mellor

, Schultz

, Przybylak

, et al. Read-across for rat oral gavage repeated-dose toxicity for short-chain mono-alkylphenols: A case study. Comput Toxicol, 2017:2; 1–11.

42.

Schultz

, Przybylak

, Richarz

A-N

, et al. Read-across of 90-day rat oral repeated-dose toxicity: A case study for selected n-alkanols. Comput Toxicol, 2017:2; 12–19.

43.

Schultz

, Przybylak

, Richarz

A-N

, et al. Read-across of 90-day rat oral repeated-dose toxicity: A case study for selected 2-alkyl-1-alkanols. Comput Toxicol, 2017:2; 28–38.

44.

Enoch

, Cronin

MTD

. A review of the electrophilic reaction chemistry involved in covalent DNA binding. Crit Rev Toxicol, 2010:40; 728–748.

45.

Enoch

, Ellison

, Schultz

, et al. A review of the electrophilic reaction chemistry involved in covalent protein binding relevant to toxicity. Crit Rev Toxicol, 2011:41; 783–802.

46.

Enoch

, Roberts

, Cronin

MTD

. Electrophilic reaction chemistry of low molecular weight respiratory sensitizers. Chem Res Toxicol, 2009:22; 1447–1453.

47.

Schwöbel

JAH

, Madden

, Cronin

MTD

. Application of a computational model for Michael addition reactivity in the prediction of toxicity to Tetrahymena pyriformis. Chemosphere, 2011:85; 1066–1074.

48.

Schwöbel

JAH

, Madden

, Cronin

MTD

. Examination of Michael addition reactivity towards glutathione by transition-state calculations. SAR QSAR Environ Res, 2010:21; 693–710.

49.

Schwöbel

JAH

, Wondrousch

, Koleva

, et al. Prediction of Michael-type acceptor reactivity toward glutathione. Chem Res Toxicol, 2010:23; 1576–1585.

50.

Schultz

, Sinks

, Cronin

MTD

. Quinone-induced aquatic toxicity: Structure-activity relationships. Aquat Toxicol, 1997:39; 267–278.

51.

Russom

, Bradbury

, Broderius

, et al. Predicting modes of toxic action from chemical structure: Acute toxicity in the fathead minnow (Pimephales promelas). Environ Toxicol Chem, 1997:16; 948–967.

52.

Verhaar

HJM

, van Leeuwen

, Hermens

JLM

. Classifying environmental pollutants. 1. Structure-Activity Relationships for prediction of aquatic toxicity. Chemosphere, 1992:25; 471–491.

53.

Nelms

, Mellor

, Cronin

MTD

, et al. Development of an in silico profiler for mitochondrial toxicity. Chem Res Toxicol, 2015:28; 1891–1902.

54.

Sinks

, Schultz

, Hunter

. UVb-induced toxicity of PAHs: Effects of substituents and heteroatom substitution. Bull Environ Contam Toxicol, 1997:59; 1–8.

55.

Veith

, Mekenyan

, Ankley

, et al. A QSAR analysis of substituent effects on the photoinduced acute toxicity of PAHs. Chemosphere, 1995:30; 2129–2142.

56.

Mekenyan

, Ankley

, Veith

, et al. QSARs for photoinduced toxicity. 1. Acute lethality of polycyclic aromatic hydrocarbons to Daphnia magna. Chemosphere, 1994:28; 567–582.

57.

Al Sharif

, Tsakovska

, Pajeva

, et al. The application of molecular modelling in the safety assessment of chemicals: A case study on ligand-dependent PPARγ dysregulation. Toxicology, 2017 (In Press). DOI: https://doi.org/10.1016/j.tox.2016.01.009

58.

Tsakovska

, Al Sharif

, Alov

, et al. Molecular modelling study of the PPARγ receptor in relation to the mode of action/Adverse Outcome Pathway framework for liver steatosis. Int J Mol Sci, 2014:15; 7651–7666.

59.

, Fisher

, Naciff

, et al. Framework for identifying chemicals with structural features associated with the potential to act as developmental or reproductive toxicants. Chem Res Toxicol, 2013:26; 1840–1861.

60.

Asako

, Uesawa

. High-performance prediction of human estrogen receptor agonists based on chemical structures. Molecules, 2017:22; 675.

61.

Feswick

, Loughery

, Isaacs

, et al. Molecular initiating events of the intersex phenotype: Low-dose exposure to 17 α-ethinylestradiol rapidly regulates molecular networks associated with gonad differentiation in the adult fathead minnow testis. Aquat Toxicol, 2016:181; 46–56.

62.

Russom

, LaLone

, Villeneuve

, et al. Development of an Adverse Outcome Pathway for acetylcholinesterase inhibition leading to acute mortality. Environ Toxicol Chem, 2014:33; 2157–2169.

63.

Lee

, Barron

. A mechanism-based 3D-QSAR approach for classification and prediction of acetylcholinesterase inhibitory potency of organophosphate and carbamate analogs. J Comput Aided Mol Des, 2016:30; 347–363.

64.

Bermudez-Saldana

, Cronin

MTD

. Quantitative structure-activity relationships for the toxicity of organophosphorus and carbamate pesticides to the rainbow trout Onchorhyncus mykiss. Pest Man Sci, 2006:62; 819–831.

65.

Könemann

. Quantitative structure-activity relationships in fish toxicity studies. Part 1: relationship for 50 industrial pollutants. Toxicology, 1981:19; 209–221.

66.

Cronin

MTD

. (Q)SARs to predict environmental toxicities: Current status and future needs. Environ Sci Process Impacts, 2017:19; 213–220.

67.

Antczak

, White

, Giri

, et al. Systems biology approach reveals a calcium-dependent mechanism for basal toxicity in Daphnia magna. Environ Sci Technol, 2015:49; 11132–11140.

68.

Przybylak

, Alzahrani

, Cronin

MTD

. How does the quality of phospholipidosis data influence the predictivity of structural alerts?. J Chem Inf Model, 2014:54; 2224–2232.

69.

Przybylak

, Cronin

MTD

. In silico studies of the relationship between chemical structure and drug induced phospholipidosis. Mol Inform, 2011:30; 415–429.

70.

Hulzebos

, Walker

, Gerner

, et al. Use of structural alerts to develop rules for identifying chemical substances with skin irritation or skin corrosion potential. QSAR Comb Sci, 2005:24; 332–342.

71.

Gerner

, Zinke

, Graetschel

, et al. Development of a decision support system for the introduction of alternative methods into local irritancy/corrosivity testing strategies. Creation of fundamental rules for a decision support system. Altern Lab Anim, 2000:28; 665–698.

72.

Worth

, Cronin

MTD

. The use of pH measurements to predict the potential of chemicals to cause dermal and ocular toxicity. Toxicology, 2001:169; 119–131.

73.

Patlewicz

, Rodford

, Ellis

, et al. A QSAR model for the eye irritation of cationic surfactants. Toxicol in vitro, 2000:14; 79–84.

74.

Hooper

, Ankley

, Cristol

, et al. Interactions between chemical and climate stressors: A role for mechanistic toxicology in assessing climate change risks. Environ Toxicol Chem, 2013:32; 32–48.

75.

AOP Wiki. https://aopwiki.org (last accessed Aug. 22, 2017).

76.

AOP Knowledge Base. http://aopkb.org (last accessed Aug. 22, 2017).

77.

van Wezel

, Opperhuizen

. Narcosis due to environmental pollutants in aquatic organisms—Residue-based toxicity, mechanisms, and membrane burdens. Crit Rev Toxicol, 1995:25; 255–279.

78.

Vinken

, Blaauboer

. In vitro testing of basal cytotoxicity: Establishment of an adverse outcome pathway from chemical insult to cell death. Toxicol in vitro, 2017:39; 104–110.

79.

Enoch

, Madden

, Cronin

MTD

. Identification of mechanisms of toxic action for skin sensitisation using a SMARTS pattern based approach. SAR QSAR Environ Res, 2008:19; 555–578.

80.

Dimitrov

, Diderich

, Sobanski

, et al. QSAR Toolbox—Workflow and major functionalities. SAR QSAR Environ Res, 2016:27; 203–219.

81.

Benigni

, Bossa

, Tcheremenskaia

. A data-based exploration of the Adverse Outcome Pathway for skin sensitization points to the necessary requirements for its prediction with alternative methods. Reg Toxicol Pharmacol, 2016:78; 45–52.

82.

Strickland

, Zang

, Paris

, et al. Multivariate models for prediction of human skin sensitization hazard. J Appl Toxicol, 2017:37; 347–360.

83.

Yang

, Tarkhov

, Marusczyk

, et al. New publicly available chemical query language, CSRML, to support chemotype representations for application to data mining and modeling. J Chem Inf Model, 2015:55; 510–528.

84.

Mansouri

, Abdelaziz

, Rybacka

, et al. CERAPP: Collaborative Estrogen Receptor Activity Prediction Project. Environ Health Persp, 2016:124; 1023–1033.

85.

Natsch

, Emter

. Reaction chemistry to characterize the molecular initiating event in skin sensitization: A journey to be continued. Chem Res Toxicol, 2017:30; 315–331.

86.

Ives

, Campia

, Wang

R-L

, et al. Creating a structured AOP knowledgebase via ontology-based annotations. Appl in Vitro Toxicol, 2017:3; 298–311.

87.

Effectopedia. http://effectopedia.org (last accessed Aug. 22, 2017).

88.

Patlewicz

, Kuseva

, Kesoya

, et al. Towards AOP application—Implementation of an integrated approach to testing and assessment (IATA) into a pipeline tool for skin sensitization. Reg Toxicol Pharmacol, 2014:69; 529–545.

89.

Conolly

, Ankley

, Cheng

, et al. Quantitative Adverse Outcome Pathways and their application to predictive toxicology. Environ Sci Technol, 2017:51; 4661–4672.

90.

Obiol-Pardo

, Gomis-Tena

, Sanz

, et al. A multiscale simulation system for the prediction of drug-induced cardiotoxicity. J Chem Inf Model, 2011:51; 483–492.

91.

Ribay

, Kim

, Wang

, et al. Predictive modelling of estrogen receptor binding agents using advanced cheminformatics tools and massive public data. Front Environ Sci, 2016:4; 12.

92.

Allen

TEH

, Liggi

, Goodman

, et al. Using Molecular Initiating Events to generate 2D structure–activity relationships for toxicity screening. Chem Res Toxicol, 2016:29; 1611–1627.

93.

Oki

, Nelms

, Bell

, et al. Accelerating Adverse Outcome Pathway development using publicly available data sources. Curr Environ Health Rep, 2016:3; 53–63.

94.

Huang

, Xia

, Sakamuru

, et al. Modelling the Tox21 10K chemical profiles for in vivo toxicity prediction and mechanism characterization. Nat Commun, 2016:7; 10425.

95.

Bell

, Angrish

, Wood

, et al. Integrating publicly available data to generate computationally predicted Adverse Outcome Pathways for fatty liver. Toxicol Sci, 2016:150; 510–520.

96.

Madden

. In silico. approaches for predicting ADME properties. In: Recent Advances in QSAR Studies: Methods and Applications, . T Puzyn, J Leszczynski, MTD Cronin (eds); pp. 283–304. Dordrecht, Heidelberg, London, New York: Springer; 2010.

97.

Piechota

, Cronin

MTD

, Hewitt

, et al. Pragmatic approaches to using computational methods to predict xenobiotic metabolism. J Chem Inf Model, 2013:53; 1282–1293.

98.

Cronin

MTD

, Dearden

, Dobbs

. QSAR studies of comparative toxicity in aquatic organisms. Sci Tot Environ, 1991; 109/110:431–439.

99.

Jaworska

, Hunter

, Schultz

. Quantitative structure-toxicity relationships and volume fraction analyses for selected esters. Arch Environ Contam Toxicol, 1995:29; 86–93.

100.

Koleva

, Madden

, Cronin

MTD

. Formation of categories from structure-activity relationships to allow read-across for risk assessment: Toxicity of alpha, beta-unsaturated carbonyl compounds. Chem Res Toxicol, 2008:21; 2300–2312.

101.

LaLone

, Villeneuve

, Burgoon

, et al. Molecular target sequence similarity as a basis for species extrapolation to assess the ecological risk of chemicals with known modes of action. Aquat Toxicol, 2013:144; 141–154.

102.

LaLone

, Villeneuve

, Lyons

, et al. Sequence Alignment to Predict Across Species Susceptibility (SeqAPASS): A web-based tool for addressing the challenges of cross-species extrapolation of chemical toxicity. Toxicol Sci, 2016:153; 228–245.

103.

Phillips

, Leonard

, Grulke

, et al. A workflow to investigate exposure and pharmacokinetic influences on high-throughput in vitro chemical screening based on adverse outcome pathways. Environ Health Perspect, 2016:124; 53–60.

104.

Teeguarden

, Tan

, Edwards

, et al. Completing the link between exposure science and toxicology for improved environmental health decision making: The Aggregate Exposure Pathway framework. Environ Sci Technol, 2016:50; 4579–4586.

105.

Lipinski

, Lombardo

, Dominy

, et al. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv Drug Del Rev, 2001:46; 3–26.

106.

Lipinski

. Rule of five in 2015 and beyond: Target and ligand structural limitations, ligand chemistry structure and drug discovery project decisions. Adv Drug Del Rev, 2016:101; 34–41.

107.

Ates

, Steinmetz

, Doktorova

, et al. Linking existing in vitro dermal absorption data to physicochemical properties: Contribution to the design of a weight-of-evidence approach for the safety evaluation of cosmetic ingredients with low dermal bioavailability. Reg Toxicol Pharmacol, 2016:76; 74–78.

108.

Tsakovska

, Pajeva

, Al Sharif

, et al. Quantitative structure-skin permeability relationships. Toxicology, 2017; 387:27–42.

109.

Morales

, Montoto

, Fagiolino

, et al. Current state and future perspectives in QSAR models to predict blood-brain barrier penetration in central nervous system drug R&D. Mini Rev Med Chem, 2017:17; 247–257.

110.

Worth

, Cronin

MTD

. Structure-permeability relationships for transcorneal penetration. Altern Lab Anim, 2001:28; 403–413.

111.

El-Masri

, Kleinstreuer

, Hines

, et al. Integration of life-stage physiologically based pharmacokinetic models with Adverse Outcome Pathways and environmental exposure models to screen for environmental hazards. Toxicol Sci, 2016:152; 230–243.

112.

Blaauboer

. The integration of data on physico-chemical properties, in vitro-derived toxicity data and physiologically based kinetic and dynamic modelling as a tool in hazard and risk assessment. A commentary. Toxicol Lett, 2003:138; 161–171.

113.

Patlewicz

, Ball

, Boogaard

, et al. Building scientific confidence in the development and evaluation of read-across. Reg Toxicol Pharmacol, 2015:72; 117–133.

114.

Przybylak

, Schultz

. Informing chemical categories through the development of Adverse Outcome Pathways. In: Chemical Toxicity Prediction: Category Formation and Read-Across. MTD Cronin, JC Madden, SJ Enoch, et al. (eds); pp. 44–71. Cambridge: Royal Society of Chemistry; 2013.

115.

Enoch

, Cronin

MTD

, Ellison

. The use of a chemistry-based profiler for covalent DNA binding in the development of chemical categories for read-across for genotoxicity. Altern Lab Anim, 2011:39; 131–145.

116.

Hewitt

, Enoch

, Madden

, et al. Hepatotoxicity: A scheme for generating chemical categories for read-across, structural alerts and insights into mechanism(s) of action. Crit Rev Toxicol, 2013:43; 537–558.

117.

Yamada

, Tanaka

, Hasegawa

, et al. Development of a category approach to predict the testicular toxicity of chemical substances structurally related to ethylene glycol methyl ether. Reg Toxicol Pharmacol, 2014:70; 711–719.

118.

Enoch

, Schultz

, Cronin

MTD

. The definition of the applicability domain relevant to skin sensitization for the aromatic nucleophilic substitution mechanism. SAR QSAR Environ Res, 2012:23; 649–663.

119.

Enoch

, Cronin

MTD

, Schultz

. The definition of the toxicologically relevant applicability domain for the SNAr reaction for substituted pyridines and pyrimidines. SAR QSAR Environ Res, 2013:24; 651–658.

120.

Schultz

, Cronin

MTD

. Lessons learned from read-across case studies for repeated-dose toxicity. Reg Toxicol Pharmacol, 2017:88; 185–191.