Expanding the Computational Drug Repositioning Toolbox

Abstract

Repositioning of previously approved drugs is a promising methodology because it lowers the cost and duration of the drug development pipeline and reduces the likelihood of unforeseen adverse events. Computational repositioning is especially appealing due to the ability to rapidly and scalably screen candidates in silico. However, despite increasing interest in repositioning from both academia and industry, existing computational methods are limited in their generalizability and reproducibility. We hypothesize that by expanding the scope of data accessible for repositioning, while at the same time improving the quality of analytic validation for new methods, repositioning studies can make higher quality predictions of promising repositioning candidates. Here we describe three methods that expand the types of data that can be leveraged for computational repositioning: (1) ksRepo, which expands the scope of gene-based repositioning and allows for the use of any ‘omics modality for disease profiling, (2) MeSHDD, for which we developed a novel drug-drug similarity metric based on overlap between drugs and medical subject heading (MeSH) terms in the biomedical literature, and (3) a novel method for quantitative trait-based repositioning using deep cross-sectional phenotyping studies that explicitly addresses sources of confounding. We also describe our work to survey and improve reproducibility in the repositioning field, and introduce a new standard database, repoDB, that contains over 10,000 drug-disease pairs, including examples of both approved and failed pairs. Together, the methods and resources described here represent an expansion of the computational drug repositioning toolbox, as well as a first step towards enabling reproducible validation of drug repositioning methods.