Constructing Adverse Outcome Pathways: a Demonstration of an Ontology-based Semantics Mapping Approach
Citation:
Wang, R., C. Ives, N. Oki, M. Nelms, AND S. Edwards. Constructing Adverse Outcome Pathways: a Demonstration of an Ontology-based Semantics Mapping Approach. 2017 SOT Annual Meeting, Baltimore, MD, March 12 - 16, 2017.
Impact/Purpose:
Developing a novel phenomics approach to further extend the molecular findings from Agency ToxCast program to phenotypes of greater regulatory relevance could potentially yield a large number of putative adverse outcome pathways, a conceptual framework facilitating future assessment of chemical risks. This presentation summarizes our implementation of this approach and initial evaluation of its effectiveness in this regard.
Description:
Adverse outcome pathway (AOP) provides a conceptual framework to evaluate and integrate chemical toxicity and its effects across the levels of biological organization. As such, it is essential to develop a resource-efficient and effective approach to extend molecular initiating events (MIEs) of chemicals to their downstream phenotypes of a greater regulatory relevance. A number of ongoing public phenomics (high throughput phenotyping) efforts have been generating abundant phenotypic data annotated with ontology terms. These phenotypes can be analyzed semantically and linked to MIEs of interest, all in the context of a knowledge base integrated from a variety of ontologies for various species and knowledge domains. In such analyses, two phenotypic profiles (PPs; anchored by genes or diseases) each characterized by multiple ontology terms are compared for their semantic similarities within a common ontology graph, but across boundaries of species and knowledge domains. Taking advantage of publicly available ontologies and software tool kits, we have implemented an OS-Mapping (Ontology-based Semantics Mapping) approach as a Java application, and constructed a network of 19383 PPs as nodes with edges weighed by their pairwise semantic similarity scores. Individual PPs were assembled from public phenomics data. Out of possible 1.87×108 pairwise connections among these nodes, about 71% of them have similarity scores between 0.2 and the maximum possible of 1.0. This network enables a MIE of interest to be mapped to multiple phenotypes structured ontologically. The October, 2015 release of the ToxCast/Tox21 program contains screening data of 375 unique target genes or proteins (out of 450 planned) by 360 unique assays for over 9000 chemicals. More than 6500 chemicals were determined to be active on at least one target. Among the 450 targets, 207 (409 if orthologs considered) could be mapped to the nodes in our phenotypic network, thus connecting numerous screened chemicals to their phenotypes. Examinations of selected MIEs and their linked phenotypes found strong evidence in literature supporting their associations, suggesting that OS-Mapping is a sound approach for developing putative AOPs for further evaluations.
