Development of a network of carcinogenicity adverse outcome pathways and its employment as an evidence framework for safety assessment

The traditional paradigm for safety assessment of chemicals for their carcinogenic potential to humans relies heavily on a battery of well-established genotoxicity tests, usually followed up by long-term, high-dose rodent studies. There are a variety of problems with this approach, not least that the rodent may not always be the best model to predict toxicity in humans. Consequently, new approach methodologies (NAMs) are being developed to replace or enhance predictions coming from the existing assays. However, a combination of the data arising from NAMs is likely to be required to improve upon the current paradigm, and consequently a framework is needed to combine evidence in a meaningful way. Adverse outcome pathways (AOPs) represent an ideal construct on which to organize this evidence. In this work, a data structure outlined previously was used to capture AOPs and evidence relating to carcinogenicity. Knowledge held within the predictive system Derek Nexus was extracted, built upon, and arranged into a coherent network containing 37 AOPs. 60 assays and 351 in silico alerts were then associated with KEs in this network, and it was brought to life by associating data and contextualizing evidence and predictions for over 13,400 compounds. Initial investigations into using the network to view knowledge and reason between evidence in different ways were made. Organizing knowledge and evidence in this way provides a flexible framework on which to carry out more consistent and meaningful carcinogenicity safety assessments in many different contexts.

Beyond adverse outcome pathways: making toxicity predictions from event networks, SAR models, data and knowledge.

Adverse outcome pathways have shown themselves to be useful ways of understanding and expressing knowledge about sequences of events that lead to adverse outcomes (AOs) such as toxicity. In this paper we use the building blocks of adverse outcome pathways-namely key events (KEs) and key event relationships-to construct networks which can be used to make predictions of the likelihood of AOs. The networks of KEs are augmented by data from and knowledge about assays as well as by structure activity relationship predictions linking chemical classes to the observation of KEs. These inputs are combined within a reasoning framework to produce an information-rich display of the relevant knowledge and data and predictions of AOs both in the abstract case and for individual chemicals. Illustrative examples are given for skin sensitization, reprotoxicity and non-genotoxic carcinogenicity.

Prediction of the effect of formulation on the toxicity of chemicals.

Two approaches for the prediction of which of two vehicles will result in lower toxicity for anticancer agents are presented. Machine-learning models are developed using decision tree, random forest and partial least squares methodologies and statistical evidence is presented to demonstrate that they represent valid models. Separately, a clustering method is presented that allows the ordering of vehicles by the toxicity they show for chemically-related compounds.

A quantitative in silico model for predicting skin sensitization using a nearest neighbours approach within expert-derived structure-activity alert spaces.

Dermal contact with chemicals may lead to an inflammatory reaction known as allergic contact dermatitis. Consequently, it is important to assess new and existing chemicals for their skin sensitizing potential and to mitigate exposure accordingly. There is an urgent need to develop quantitative non-animal methods to better predict the potency of potential sensitizers, driven largely by European Union (EU) Regulation 1223/2009, which forbids the use of animal tests for cosmetic ingredients sold in the EU. A Nearest Neighbours in silico model was developed using an in-house dataset of 1096 murine local lymph node (LLNA) studies. The EC3 value (the effective concentration of the test substance producing a threefold increase in the stimulation index compared to controls) of a given chemical was predicted using the weighted average of EC3 values of up to 10 most similar compounds within the same mechanistic space (as defined by activating the same Derek skin sensitization alert). The model was validated using previously unseen internal (n = 45) and external (n = 103) data and accuracy of predictions assessed using a threefold error, fivefold error, European Centre for Ecotoxicology and Toxicology of Chemicals (ECETOC) and Globally Harmonized System of Classification and Labelling of Chemicals (GHS) classifications. In particular, the model predicts the GHS skin sensitization category of compounds well, predicting 64% of chemicals in an external test set within the correct category. Of the remaining chemicals in the previously unseen dataset, 25% were over-predicted (GHS 1A predicted: GHS 1B experimentally) and 11% were under-predicted (GHS 1B predicted: GHS 1A experimentally). Copyright © 2017 John Wiley & Sons, Ltd.

A k-Nearest Neighbours Approach Using Metabolism-related Fingerprints to Improve In Silico Metabolite Ranking.

The application of biotransformation dictionaries derived by expert evaluation of known metabolic pathways represents one approach to the prediction of both phase I and phase II xenobiotic metabolites. The ranking of metabolites generated by such dictionaries has previously been achieved through the use of qualitative reasoning rules and quantitative probability values. Using the biotransformation dictionary available in the Meteor expert system, we show that metabolite over-prediction by both of these methods can be reduced by the adoption of a k-nearest neighbours methodology in which the likelihood of a predicted biotransformation is determined based on comparison of a query chemical with structurally-similar substrates with known experimental metabolic data which activate the same biotransformation. Optimal performance was achieved when similarity was defined in terms of a combination of two fingerprints, one describing the overall profile of biotransformations a structure can potentially undergo and the other describing the local environment around the predicted site of metabolism for the particular biotransformation under consideration.

Evaluation of a statistics-based Ames mutagenicity QSAR model and interpretation of the results obtained.

The relative wealth of bacterial mutagenicity data available in the public literature means that in silico quantitative/qualitative structure activity relationship (QSAR) systems can readily be built for this endpoint. A good means of evaluating the performance of such systems is to use private unpublished data sets, which generally represent a more distinct chemical space than publicly available test sets and, as a result, provide a greater challenge to the model. However, raw performance metrics should not be the only factor considered when judging this type of software since expert interpretation of the results obtained may allow for further improvements in predictivity. Enough information should be provided by a QSAR to allow the user to make general, scientifically-based arguments in order to assess and overrule predictions when necessary. With all this in mind, we sought to validate the performance of the statistics-based in vitro bacterial mutagenicity prediction system Sarah Nexus (version 1.1) against private test data sets supplied by nine different pharmaceutical companies. The results of these evaluations were then analysed in order to identify findings presented by the model which would be useful for the user to take into consideration when interpreting the results and making their final decision about the mutagenic potential of a given compound.

Self organising hypothesis networks: a new approach for representing and structuring SAR knowledge.

BACKGROUND: Combining different sources of knowledge to build improved structure activity relationship models is not easy owing to the variety of knowledge formats and the absence of a common framework to interoperate between learning techniques. Most of the current approaches address this problem by using consensus models that operate at the prediction level. We explore the possibility to directly combine these sources at the knowledge level, with the aim to harvest potentially increased synergy at an earlier stage. Our goal is to design a general methodology to facilitate knowledge discovery and produce accurate and interpretable models. RESULTS: To combine models at the knowledge level, we propose to decouple the learning phase from the knowledge application phase using a pivot representation (lingua franca) based on the concept of hypothesis. A hypothesis is a simple and interpretable knowledge unit. Regardless of its origin, knowledge is broken down into a collection of hypotheses. These hypotheses are subsequently organised into hierarchical network. This unification permits to combine different sources of knowledge into a common formalised framework. The approach allows us to create a synergistic system between different forms of knowledge and new algorithms can be applied to leverage this unified model. This first article focuses on the general principle of the Self Organising Hypothesis Network (SOHN) approach in the context of binary classification problems along with an illustrative application to the prediction of mutagenicity. CONCLUSION: It is possible to represent knowledge in the unified form of a hypothesis network allowing interpretable predictions with performances comparable to mainstream machine learning techniques. This new approach offers the potential to combine knowledge from different sources into a common framework in which high level reasoning and meta-learning can be applied; these latter perspectives will be explored in future work.

Emerging pattern mining to aid toxicological knowledge discovery.

Knowledge-based systems for toxicity prediction are typically based on rules, known as structural alerts, that describe relationships between structural features and different toxic effects. The identification of structural features associated with toxicological activity can be a time-consuming process and often requires significant input from domain experts. Here, we describe an emerging pattern mining method for the automated identification of activating structural features in toxicity data sets that is designed to help expedite the process of alert development. We apply the contrast pattern tree mining algorithm to generate a set of emerging patterns of structural fragment descriptors. Using the emerging patterns it is possible to form hierarchical clusters of compounds that are defined by the presence of common structural features and represent distinct chemical classes. The method has been tested on a large public in vitro mutagenicity data set and a public hERG channel inhibition data set and is shown to be effective at identifying common toxic features and recognizable classes of toxicants. We also describe how knowledge developers can use emerging patterns to improve the specificity and sensitivity of an existing expert system.

Feature combination networks for the interpretation of statistical machine learning models: application to Ames mutagenicity.

BACKGROUND: A new algorithm has been developed to enable the interpretation of black box models. The developed algorithm is agnostic to learning algorithm and open to all structural based descriptors such as fragments, keys and hashed fingerprints. The algorithm has provided meaningful interpretation of Ames mutagenicity predictions from both random forest and support vector machine models built on a variety of structural fingerprints.A fragmentation algorithm is utilised to investigate the model's behaviour on specific substructures present in the query. An output is formulated summarising causes of activation and deactivation. The algorithm is able to identify multiple causes of activation or deactivation in addition to identifying localised deactivations where the prediction for the query is active overall. No loss in performance is seen as there is no change in the prediction; the interpretation is produced directly on the model's behaviour for the specific query. RESULTS: Models have been built using multiple learning algorithms including support vector machine and random forest. The models were built on public Ames mutagenicity data and a variety of fingerprint descriptors were used. These models produced a good performance in both internal and external validation with accuracies around 82%. The models were used to evaluate the interpretation algorithm. Interpretation was revealed that links closely with understood mechanisms for Ames mutagenicity. CONCLUSION: This methodology allows for a greater utilisation of the predictions made by black box models and can expedite further study based on the output for a (quantitative) structure activity model. Additionally the algorithm could be utilised for chemical dataset investigation and knowledge extraction/human SAR development.

Automating knowledge discovery for toxicity prediction using jumping emerging pattern mining.

The design of new alerts, that is, collections of structural features observed to result in toxicological activity, can be a slow process and may require significant input from toxicology and chemistry experts. A method has therefore been developed to help automate alert identification by mining descriptions of activating structural features directly from toxicity data sets. The method is based on jumping emerging pattern mining which is applied to a set of toxic and nontoxic compounds that are represented using atom pair descriptors. Using the resulting jumping emerging patterns, it is possible to cluster toxic compounds into groups defined by the presence of shared structural features and to arrange the clusters into hierarchies. The methodology has been tested on a number of data sets for Ames mutagenicity, oestrogenicity, and hERG channel inhibition end points. These tests have shown the method to be effective at clustering the data sets around minimal jumping-emerging structural patterns and finding descriptions of potentially activating structural features. Furthermore, the mined structural features have been shown to be related to some of the known alerts for all three tested end points.

Towards the creation of an international toxicology information centre.

A pilot toxicology database system has been created which is accessible on-line via the world-wide web or in-house via an intranet. It is intended to be suitable as a source of toxicological information and to support structure-activity relationship studies, and it can be searched on chemical structural and substructural as well as toxicological and physico-chemical data. Successful completion of the pilot has led to an ongoing project to develop and expand the system.

A comprehensive approach to argumentation.

A reasoning model, based on the logic of argumentation, is described. The model represents argumentation as a directed graph in which nodes and arcs can be colored using an ordinal set of weightings and in which the attributes of both nodes and arcs can be modified. It is thus able to deal with the undercutting or augmenting of arguments. Weightings can be propagated through the graph to generate unique weightings for any node or arc. The model is able to deal with contradiction. It can incorporate numerical methods and is able to handle qualitative and quantitative reasoning.

Using absolute and relative reasoning in the prediction of the potential metabolism of xenobiotics.

To be useful, a system which predicts the metabolic fate of a chemical should predict the more likely metabolites rather than every possibility. Reasoning can be used to prioritize biotransformations, but a real biochemical domain is complex and cannot be fully defined in terms of the likelihood of events. This paper describes the combined use of two models for reasoning under uncertainty in a working system, METEOR-one model deals with absolute reasoning and the second with relative reasoning.

Using argumentation for absolute reasoning about the potential toxicity of chemicals.

The application of a new argumentation model is illustrated by reference to DEREK for Windows, a knowledge-based expert system for the prediction of the toxicity of chemicals. Examples demonstrate various aspects of the model such as the undercutting of arguments, the resolution of multiple arguments about the same proposition, and the propagation of arguments along a chain of reasoning.