RePurpose: An LLM-Assisted, Source-Traceable Drug Repurposing Tool with Retrospective Validation on Clemastine in Multiple Sclerosis

RePurpose is a web-based tool that assists in the repurposing of drugs by utilizing information from various data sources such as Open Targets, ClinicalTrials.gov, UniProt, and PubMed. It allows users to input a topic in plain language and generates a list of candidate drugs based on evidence from the aforementioned databases. The tool filters the candidate drugs based on whether they modulate all the targets implicated in the goal in the corresponding direction.

Each candidate drug is accompanied by a source that provides evidence, which could be a structured database, a clinical trial, or a proposal from a large language model. The results are ranked, with options to adjust the order from well-studied to less known drugs.

The tool was tested retrospectively with clemastine, a drug that was initially considered as a remyelinating agent in multiple sclerosis (MS). The results showed that RePurpose successfully retrieved clemastine in all replicates of mechanism-worded queries, but it was not found in any replicates of disease-worded queries. Moreover, even with the language model enabled, clemastine was not retrieved in any structured queries.

The ranked outputs of clemastine, as generated by RePurpose, were supported by the evidence but did not account for any later evidence of harm in progressive MS. This evidence could only be accessed through an on-demand literature search.

The findings suggest that RePurpose may be more effective in tracing evidence around a hypothesis rather than generating new hypotheses. RePurpose Pharma, the tool, is freely accessible at https://www.repurposepharma.com/. For those interested in the application, code, and more information, they can visit the GitHub repository at https://github.com/nayagamnitin-hub/RePurpose-Pharma.