The AOPOntology: A Semantic Artificial Intelligence Tool for Predictive Toxicology

Abstract

IntroductionToxicology needs artificial intelligence tools that can automate the prediction of toxicity. Today we are at an interesting nexus. We have thousands of chemicals in the environment that lack regulatory thresholds for determining risk. New high throughput in vitro testing methods are becoming available to test these chemicals. Causal Adverse Outcome Pathway Networks (CAOPN) are emerging that will allow us to make predictions based on perturbations of specific key events within the network. The AOPOntology was developed as infrastructure for this nexus, providing the ability to model and marry the data from the in vitro tests for the thousands of chemicals and place them within the CAOPN framework to facilitate adverse outcome predictions.\n\nMaterials and MethodsThe AOPN is a functional specialized ontology that creates classes that model biological pathways and CAOPNs. Adverse outcome predictions are based on mathematical determinations of key events that are sufficient to infer adverse outcomes will occur, or biological information. These sufficiency relationships are captured in the AOPOntology and used by the semantic reasoners to make predictions.\n\nResultsThe AOPOntology version 1.0 architecture is in place, and a CAOPN for steatosis demonstrates how causal network theory is used to make predictions. The AOPOntology is available at https://github.com/DataSciBurgoon/aop-ontology.\n\nDiscussionThe AOPOntology is a knowledge base for CAOPNs that one can use to make predictions about a chemicals potential toxicity using in vitro high throughput and other assays.\n\nConclusionsUsing CAOPNs and causal network theory one is able to predict potential toxicity for chemicals using in vitro high throughput and various high content screens.