Predicting Adverse Drug Effects from Literature Assertions that Link Drugs to Targets and Targets to Effects
Mary K. La · Carolina Digital Repository (University of North Carolina at Chapel Hill) · 2019
Adverse drug effects (ADEs) are a major reason for drug candidate failure in clinical trials; thus, it is critical to predict possible ADEs in the early stages of drug discovery. In this study, cheminformatics, bioinformatics, and data mining approaches were employed to integrate and analyze publicly-available pharmacological and clinical data with the goal of inferring novel associations between drugs, targets, and ADEs. A new database was created that integrated experimental drug-target binding data and known associations between drugs (7448 unique instances), targets (1280), and ADEs (4492) expressed as assertions found in the literature. Unreported associations between drugs, targets, and ADEs were inferred, and inferences were prioritized as testable hypotheses. As a proof of concept, an association was identified between paroxetine and thrombocytopenic purpura using a focused subset of ~47K top-ranked inferences published prior to the first reports confirming this association in 2013. Given the increasing costs of bringing new drug entities to market, there is a strong need for cost-effective methods of identifying potential adverse effects of a drug candidate early on in the development process. The workflow presented here, based on free-access databases and an association-based inference scheme, has provided chemical-ADE inferences that have been validated post-hoc in literature case reports. With refinement of prioritization schemes for the generated chemical-ADE inferences, this workflow may provide an effective computational method for the early detection of drug candidate ADEs.