Summary Drug-drug interactions (DDIs) and pharmacogenetics (PG) are leading causes of adverse drug events (ADEs), with one in four patients experiencing ADEs attributable to DDIs or PG. However, despite the intrinsic connection of their pharmacological mechanisms, DDI and PG are often studied separately. There is a significant need for more efficient and effective translational from DDI to PG research, and newly developed machine-learning (ML) and artificial-intelligence (AI) methods have made such research feasible. In our recent DDI knowledge-discovery study of 25 million PubMed abstracts, we used ML and natural-language-processing analyses for the first time to identify 986 DDI pairs with overlapping pharmacokinetic mechanisms and clinical evidence, from which we generated 137 new PG hypotheses regarding CYP2D6 and CYP3A. In this grant proposal, we will develop novel ML methods, including active learning that will allow human annotator involvement and knowledge base reasoning that relies on logical rules to represent pharmacological mechanisms. This proposal has three aims: (1) to develop an active-learning approach to perform DDI and PG information retrieval analysis from the literature; (2) to develop a joint information-extraction and knowledge- base-reasoning approach to perform DDI and PG information extraction analysis from the literature; and (3) (a) to examine whether CYP3A/CYP2C19 genetic polymorphisms are associated with omeprazole-induced myopathy, and (b) to develop a prioritization scheme to examine new PG hypotheses generated from the literature-based discovery analyses from Aims 1 and 2 using Vanderbilt University’s BioVU biobank. These PG findings will provide a valuable resource for the wider scientific community for potential prospective studies and contribute significantly to the improvement of precision medicine and clinical care.

Narrative In order to discover and generate new pharmacogenetic hypotheses from drug interaction evidences published in the scientific literature, we will develop novel machine learning methods, including active learning allowing human involvement and knowledge base reasoning. These new pharmacogenetics hypotheses will be tested in a biobank. These pharmacogenetics findings will contribute significantly towards precision medicine and improve clinical care.