| We investigated the use of semantics to aid data integration and predictive model development. Toxicity data from multiple sources was converted into a Semantic Web Linked Data resource. Datasources included the U.S Environmental Protection Agency ToxCast program [1], the Comparative Toxicogenomics Database [2], and the U.S National Library of Medicine TOXNET archives [3]. The resource was used to develop predictive toxicology models represented as OWL-encoded ontologies. The use of OWL enabled automatic classification of linked data, model integration, and logical interpretation of results. A framework was developed to represent molecular mechanisms of action allowing reasoning and semantic querying answering. Finally, we investigated how this framework could be used to infer novel features for machine learning input. The completed work shows how formal knowledge representation can be used to improve integration of toxicology information and development of predictive models. |
