Zero-shot drug repurposing with geometric deep learning and clinician centered design

Historically, drug repurposing – identifying new therapeutic uses for approved drugs – has been attributed to serendipity. While recent advances have leveraged knowledge graphs and deep learning to identify potential therapeutic candidates, their clinical utility remains limited because they focus on diseases with available existing treatments and rich molecular knowledge. Here, we introduce TXGNN, a geometric deep learning approach designed for “zero-shot” drug repurposing, identifying therapeutic candidates even for diseases with no existing medicines. Trained on a medical knowledge graph, TXGNN utilizes a graph neural network and metric-learning module to rank therapeutic candidates as potential indications and contraindications across 17,080 diseases. When benchmarked against eight methods, TXGNN significantly improves prediction accuracy for indications by 49.2% and contraindications by 35.1% under stringent zero-shot evaluation. To facilitate interpretation and analysis of the model’s predictions, TXGNN’s Explainer module offers transparent insights into the multi-hop paths that form TXGNN’s predictive rationale. Our pilot human evaluation of TXGNN’s Explainer showed that TXGNN’s novel predictions and explanations perform encouragingly on multiple axes of model performance beyond accuracy. Many of TXGNN’s novel predictions are aligned with off-label prescriptions made by clinicians within a large healthcare system, affirming their potential clinical utility. TXGNN provides drug repurposing predictions that are more accurate than existing methods, are consistent with off-label prescription decisions made by clinicians, and can be investigated by human experts through multi-hop interpretable explanations. ### Competing Interest Statement The authors have declared no competing interest. ### Funding Statement K.H., P.C., and M.Z.~gratefully acknowledge the support of NIH R01-HD108794, US DoD FA8702-15-D-0001, awards from Harvard Data Science Initiative, Amazon Faculty Research, Google Research Scholar Program, Bayer Early Excellence in Science, AstraZeneca Research, Roche Alliance with Distinguished Scientists, and Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University. P.C. was supported, in part, by the Harvard Summer Institute in Biomedical Informatics. Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the funders. ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: All clinical and electronic medical record data were deidentified, and the Institutional Review Board at Mount Sinai, New York City, U.S., approved the study. I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes TXGNN’s website is at . The knowledge graph dataset is available at Harvard Dataverse under a persistent identifier . All clinical and electronic medical record data were deidentified, and the Institutional Review Board at Mount Sinai, New York City, U.S., approved the study.