Computational Drug Repositioning: From Data to Therapeutics

Traditionally, most drugs have been discovered using phenotypic or target-based screens. Subsequently, their indications are often expanded on the basis of clinical observations, providing additional benefit to patients. This review highlights computational techniques for systematic analysis of transcriptomics (Connectivity Map, CMap), side effects, and genetics (genome-wide association study, GWAS) data to generate new hypotheses for additional indications. We also discuss data domains such as electronic health records (EHRs) and phenotypic screening that we consider promising for novel computational repositioning methods.

Since the advent of the genomic era, most of the searches for new drugs have begun with the concept of a single target that acts through a specific mechanism. In some cases, the target is genetically linked to the disease; in others, mechanistic hypotheses lead to a biochemical assay screen of the target, and the resulting tool compound is further evaluated in relevant model systems. Other searches begin with a phenotypic screen in which the model system itself is screened for efficacious compounds. In all these cases, developers then optimize the lead compound, hoping to avoid side effects due to either off-target binding or unanticipated physiologic roles of the intended target.1 Between 1999 and 2008, this process resulted in only 50 first-in-class small-molecule agents being approved by the US Food and Drug Administration. Of these, 17 were identified as arising from target-based discovery methods and 28 from phenotypic discovery methods.2

Drug repositioning (also referred to as repurposing) has long been a necessary strategy of drug development3,4,5 because it can renew a failed drug or expand the number of indications for a successful one.6 Figure 1 highlights the differences in the time lines of repositioning as compared with those of traditional drug discovery methods. Potentially, repositioning can reduce the traditional time line of 10–17 years and make drugs available for use in patients in 3–12 years.3 Computational repositioning is the process of designing and validating automated workflows that can generate hypotheses for new indications for a drug candidate. The potential for computational repositioning is high, given that a systematic process can incorporate prioritization information that can accelerate time lines even further. Indications that lend themselves to a quick proof-of-concept or experimental-medicine study can be taken into account, and small clinical studies can be rapidly initiated for compounds for which safety data in human patients are already available.