Date of Award

Spring 1-1-2015

Document Type


Degree Name

Doctor of Philosophy (PhD)


Computer Science

First Advisor

James H Martin

Second Advisor

Elizabeth White

Third Advisor

Richard Osborne

Fourth Advisor

Ken Anderson

Fifth Advisor

Michel Dumontier


Drug-drug interactions (DDIs) constitute a major cause of adverse drug events (ADEs), which may result in morbidity, mortality, and increased healthcare expenditures. As the role of drug therapy continues to expand and polypharmacy becomes more common, the prevalence of significant DDIs also increases, potentially limiting the therapeutic benefits of medication therapy. While existing DDI research largely focuses on single levels of DDI, a method that incorporates clinical, pharmacological, and physiological (genetic) factors can offer an improved approach to identifying and characterizing potential DDIs. Current limitations to efficient DDI characterization that can affect DDI prevention include: limited availability of clinical studies, shortcomings within study designs, limited accessibility of drug interaction information due to storage in disparate sources, and omissions in DDI reporting within information sources, especially their mechanisms of interaction and clinical significance.

This thesis presents the novel Drug-drug interaction Discovery and Demystification (D3) pharmacovigilance system, which employs Big Data mining and Semantic Web technologies to predict potential DDIs, assess their clinical significance, and describe the mechanism(s) responsible for predicted interactions. By integrating drug information from a variety of trusted biomedical sources into a coherent, comprehensive DDI knowledge base, D3 leverages the power of big data to construct inference-based predictions to support probabilistic identification, validation, and mechanistic classification of potential DDIs. To qualify the effectiveness of D3, an unbiased benchmark was constructed to characterize the likelihood determinations of D3’s predictions against both DDIs reported by its own knowledge sources as well as DDIs reported for five commonly prescribed medications with high propensities for interaction by Micromedex, a respected commercial knowledge source. D3 demonstrates a 93.4% recall rate against DDIs from its own knowledge sources and performs comparably within the margin of error for DDIs reported by Micromedex. The application of D3 to DDIs predicted by five publicly available pharmacovigilance systems indicates that these systems appear to be vastly over-stating the number of DDIs. These results indicate the potential of D3 as an investigative tool for clinicians and researchers to gain some foresight into the likelihoods and causes of potential interactions.