Ontology-based Semantic Mapping of Adverse Outcome Pathways
                                                              Rong-Lin Wanga
    a
        Exposure Methods and Measurements Division, National Exposure Research Laboratory, US EPA, Cincinnati, OH 45268, USA


Abstract                                                                       current reliance on the manually defined key event relationships,
                                                                               and aid the future development of additional AOPs.
Most of the nearly 85,000 chemicals currently listed in the US
                                                                               Keywords:
TSCA (Toxic Substances Control Act) inventory are not
characterized toxicologically. A paradigm shift has been well                  adverse outcome pathway, chemical toxicity, semantic analysis
underway to move away from animal toxicity tests, and towards
more resource-efficient in vitro, in silico, and short-term in vivo            References
screenings. As such, there is a great need to link toxicity
phenotypes at molecular levels to those with greater regulatory                1. Ankley, G.T., et al., 2010. Adverse outcome pathways: a
relevance at higher levels of biological organization. The                     conceptual framework to support ecotoxicology research and
framework of adverse outcome pathway (AOP) was proposed to                     risk assessment. Environ. Toxicol. Chem. 29 (3), 730–741.
address this need (1), and has been increasingly adopted in                    https://doi.org/10.1002/etc.34.
recent years to organize toxicity information along such a                     2. Wang, R-L., Edwards, S., Ives, C., 2019. Ontology-based
biological hierarchy. Many AOPs, each consisting of a                          semantic mapping of chemical toxicities. Toxicology, 412:89-
molecular initiating event, several key events, and an adverse                 100.
outcome, have been developed (https://aopwiki.org/).
                                                                               3. Horridge, M., Bechhofer, S., 2011. The OWL API: a java API
With abundant phenotypic data from ongoing public phenomics                    for OWL ontologies. Semantic Web J. 2 (1), 11–21.
efforts and years of toxicity studies, ontology-based semantic                 https://doi.org/10.3233/SW-2011-0025.
mapping (OS-Mapping) offers a promising approach to bridge
the gaps between molecular phenotypes and traditional                          4. Harispe, S., 2014. The semantic measures library and toolkit:
endpoints provided by animal tests. To study the applications of               fast computation of semantic similarity and relatedness using
OS-Mapping in evaluating existing AOPs and aiding their future                 biomedical ontologies. Bioinformatics 30 (5), 740–742.
development, over 1100 key events belonging to more than 200                   https://doi.org/10.1093/bioinformatics/btt581.
AOPs were annotated by using entity-quality (EQ) statements.
Also included in the study were toxicity responses previously
annotated from more than 700 exposure studies of ten chemicals
in six vertebrate species (2). Together, they were assembled into
over 200 phenotypic profiles as queries, and compared
semantically to more than 37 thousand phenotypic profiles
organized by genes, diseases, and biological pathways (KEGG,
Kyoto Encyclopedia of Genes and Genomes,
https://www.genome.jp/kegg/; Reactome, https://reactome.org/)
of human, mouse, and zebrafish. The Java application for
semantic analysis was developed in-house based on OWLAPI
(version 4.2.5)(3), several publicly available reasoners, and the
Semantic Measure Library (SML, version 0.9.4d)(4). The
analyses proved to be insightful. For example, many AOPs
appeared to be quite robust, as suggested by their respective key
events having mutual similarities significantly above
background. However, most of the key event pairs curated to be
adjacent to each other (i.e., KE X upstream biologically leads to
KE Y downstream) had similarities, ranging between zero and
one, less than 0.2. Some of the key events from different AOPs
were found to be highly similar to one another, leading to their
hosting AOPs to become substantially related too. Many AOPs
were also mapped to various genes, KEGG/Reactome pathways,
and diseases. The findings like these will help to delineate the
biology underlying these AOPs and provide some independent
evidence for their robustness. Furthermore, semantic
characterization of key events and AOPs will also provide an
approach to construct AOP networks complementary to the



  Copyright © 2019 for this paper by its author. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).