Drug-drug interactions (DDIs) are a major cause of morbidity and mortality and lead to increased health care costs 
. DDIs are responsible for nearly 3% of all hospital admissions 
and 4.8% of admissions in the elderly 
. And with new drugs entering the market at a rapid pace (35 novel drugs approved by the FDA in 2011), identification of new clinically significant drug interactions is essential. DDIs are also a common cause of medical errors, representing 3% to 5% of all inpatient medication errors 
. These numbers may actually underestimate the true public health burden of drug interactions as they reflect only well-established DDIs.
Several methodological approaches are currently used to identify and characterize new DDIs. In vitro
pharmacology experiments use intact cells (e.g. hepatocytes), microsomal protein fractions, or recombinant systems to investigate drug interaction mechanisms. The FDA provides comprehensive recommendations for in vitro
study designs, including recommended probe substrates and inhibitors for various metabolism enzymes and transporters 
. The drug interaction mechanisms and parameters obtained from these in vitro
experiments can be extrapolated to predict in vivo
changes in drug exposure. For example, a physiologically based pharmacokinetics model was developed to predict the clinical effect of mechanism based inhibition of CYP3A by clarithromycin from in vitro
. However, in vitro
experiments alone often cannot determine whether a given drug interaction will affect drug efficacy or lead to a clinically significant adverse drug reaction (ADR).
clinical pharmacology studies utilize either randomized or cross-over designs to evaluate the effect on an interaction on drug exposure. Drug exposure change serves as a biomarker for the direct DDI effect, though drug exposure change may or may not lead to clinically significant change in efficacy or ADRs. The FDA provides well-documented guidance for conducting in vivo
clinical pharmacology DDI studies 
. If well-established probe substrates and inhibitors are used, involvement of specific drug metabolism or transport pathway can be demonstrated by in vivo
clinical studies. For example, using selective probe substrates of OATPs (pravastatin) and CYP3A (midazolam) and probe inhibitors of OATPs (rifampicin) and CYP3A (itraconazole), it was shown that hepatic uptake via OATPs made the dominant contribution to the hepatic clearance of atorvastatin in an in vivo
clinical PK study 
. However, due to overlap in substrate selectivity, an in vivo
DDI study alone will often not provide mechanistic insight into the DDI.
Finally, in populo
pharmacoepidemiology studies use a population-based approach to investigate the effect of a DDI on drug efficacy and ADRs. For example, the interactions between warfarin and several antibiotics were evaluated for increased risk of gastrointestinal bleeding and hospitalization in a series of case-control and case-crossover studies using US Medicaid data 
. Indeed, epidemiological studies using large clinical datasets can identify potentially interacting drugs within a population, but these studies alone are insufficient to characterize pharmacologic mechanisms or patient-level physiologic effects.
The aforementioned in vitro
, in vivo
, and in populo
research methods are complementary in characterizing new drug-drug interactions. Yet these methods are all limited by their relatively small scale. Such studies usually focus on a few drug pairs for one or a limited number of metabolizing enzymes or transporters a time. Performing large scale screening for novel drug interactions requires higher throughput strategies. Literature mining and data mining have become powerful tools for knowledge discovery in biomedical informatics, and are particularly useful for hypothesis generation. A recent notable example in clinical pharmacology is the successful detection of novel DDIs through mining of the FDA's Adverse Event Reporting System 
. In this study, pravastatin and paroxetine were found to have a synergistic effect on increasing blood glucose. This finding was validated in three large electronic medical record (EMR) databases. While a ground-breaking success, this approach provides little evidence regarding the mechanism of the interaction.
In this paper, we present a novel approach using literature mining for screening of potential DDIs based on mechanistic properties, followed by EMR-based validation to identify those interactions that are clinically significant. We focus on clinically and statistically significant DDIs that increase the risk of myopathy.