| Finding connections between an existing drug product and its new application areas evolves to be one of the alternative and efficient strategies for new drug development. However, the identification of such connections remains highly dependent on serendipitous observation and educated guess. A forecasting informatics model that can improve data capture, integration, prediction and interpretation of potential new therapeutic indications for drugs based on integrated biomedical knowledge around drug and disease mechanisms is highly desirable. To pursue a systematic approach to the discovery of novel and inferable relationships between drugs and diseases based on mechanistic knowledge, we aim to develop a semantic infrastructure that integrates heterogeneous data from pharmacological and biological domains to allow efficient mining of non-trivial connections among biomedical and pharmacological entities across knowledge domains. In this work, we devised a Disease-Drug Correlation Ontology (DDCO), an ontological framework to integrate varied datasets extracted from pharmacological and biological domains. The DDCO, formalized in OWL, allows the integrated representation of multiple sources of ontologies, controlled vocabularies, and data schemas. We used the DDCO framework to integrate and represent a collection of data sources including DrugBank, EntrezGene, OMIM, KEGG, BioCarta, Reactome, Human Phenome, and UMLS, and constructed a comprehensive Pharmacome-Diseasome network using RDF, which represents data in conceptual graphic format. We established and validated the multiple applications using the constructed knowledge base. More importantly, we implemented graph theoretic-based network ranking algorithms onto disease-specific BioRDF to identify and prioritize drugs for new therapeutic utilities. The work is a pioneering effort in leveraging on Semantic Web principles and technologies to apply on pharmaceutical development problems from data integration, knowledge representation, and application of graph query languages and network analysis algorithms for mining drug actions and disease mechanisms. The knowledge framework and approach established in the work can be applied to many drug R&D applications, including drug repositioning and novel target prediction, to support new hypothesis generation in the context of network pharmacology, and thus facilitate solving bottleneck questions and allow scientist to distill insights for best decision-making. |
