Elucidation of bioinformatic-guided high-prospect drug repositioning candidates for DMD via Swanson linking of target-focused latent knowledge from text-mined categorical metadata

Abstract

Duchenne Muscular Dystrophy (DMD)’s complex multi-system pathophysiology, coupled with the cost-prohibitive logistics of multi-year drug screening and follow-up, has hampered the pursuit of new therapeutic approaches. Here we conducted a systematic historical and text mining-based pilot feasibility study to explore the potential of established or previously tested drugs as prospective DMD therapeutic agents. Our approach utilized a Swanson linking-inspired method to uncover meaningful yet largely hidden deep semantic connections between pharmacologically significant DMD targets and drugs developed for unrelated diseases. Specifically, we focused on molecular target-based MeSH terms and categories as high-yield bioinformatic proxies, effectively tagging relevant literature with categorical metadata. To identify promising leads, we comprehensively assembled published reports from 2011 and sampling from subsequent years. We then determined the earliest year when distinct MeSH terms or category labels of the relevant cellular target were referenced in conjunction with the drug, as well as when the pertinent target itself was first conclusively identified as holding therapeutic value for DMD. By comparing the earliest year when the drug was identifiable as a DMD treatment candidate with that of the first actual report confirming this, we computed an Index of Delayed Discovery (IDD), which serves as a metric of Swanson-linked latent knowledge. Using these findings, we identified data from previously unlinked articles subsetted via MeSH-derived Swanson linking or from target classes within the DrugBank repository. This enabled us to identify new but untested high-prospect small-molecule candidates that are of particular interest in repurposing for DMD and warrant further investigations.