Of the several thousand diseases that affect humans, only about 500 have treatments approved by the U.S. Food and Drug Administration. Even for those with approved treatments, discovering new drugs can offer alternative options that cause fewer side effects and replace drugs that are ineffective for certain patient groups. However, identifying new therapeutic opportunities for diseases with limited treatment options remains a challenge, as existing algorithms often perform poorly. Here, we leverage recent advances in geometric deep learning and human-centered AI to introduce TxGNN, a model for identifying therapeutic opportunities for diseases with limited treatment options and minimal molecular understanding. TxGNN is a graph neural network pre-trained on a comprehensive knowledge graph of 17,080 clinically-recognized diseases and 7,957 therapeutic candidates. The model can process various therapeutic tasks, such as indication and contraindication prediction, in a unified formulation. Once trained, we show that TxGNN can perform zero-shot inference on new diseases without additional parameters or fine-tuning on ground truth labels. Evaluation of TxGNN shows significant improvements over existing methods, with up to 49.2% higher accuracy in indication tasks and 35.1% higher accuracy in contraindication tasks. TxGNN can also predict therapeutic use for new drugs developed since June 2021. To facilitate interpretation and analysis of the model's predictions by clinicians, we develop a human-AI explorer for TxGNN and evaluate its usability with medical experts. Finally, we demonstrate that TxGNN's novel predictions are consistent with off-label prescription decisions made by clinicians in a large healthcare system. These label-efficient and clinician-centered learning systems pave the way for improvements for many therapeutic tasks.

The authors have declared no competing interest.

K.H., P.C., and M.Z. gratefully acknowledge the support by NSF under No.~IIS-2030459, US Air Force under No.~FA8702-15-D-0001, and awards from Harvard Data Science Initiative, Amazon Research, Bayer Early Excellence in Science, AstraZeneca Research, and Roche Alliance with Distinguished Scientists. 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.

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

The TxGNN's project website is at https://zitniklab.hms.harvard.edu/projects/TxGNN. Therapeutics-centered knowledge graph is available at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/IXA7BM under a persistent identifier https://doi.org/10.7910/DVN/IXA7BM. We have deposited the knowledge graph and all relevant intermediate files in this repository. Python implementation of the methodology developed and used in the study is available via the project website at https://zitniklab.hms.harvard.edu/projects/TxGNN. The code to reproduce results, documentation, and usage examples are at https://github.com/mims-harvard/TxGNN. To facilitate the usage of the algorithm, we developed a TxGNN Explorer, a web-based app available at http://txgnn.org to access TxGNN's predictions.

https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/IXA7BM

https://zitniklab.hms.harvard.edu/projects/TxGNN

http://txgnn.org

https://github.com/mims-harvard/TxGNN