A foundation model for clinician-centered drug repurposing
Problem
Drug repurposing AI models focus narrowly on diseases that already have drugs, leaving diseases with limited or no treatment options unaddressed.
Why it matters
Identifying new uses for approved drugs could help patients with diseases lacking existing treatments.
What the AI did
TxGNN, a graph neural network trained on a medical knowledge graph with metric learning, ranks drugs as indications and contraindications in zero-shot across 17,080 diseases.
Results
49.2% improved indication accuracy and 35.1% improved contraindication accuracy over eight benchmarked methods under zero-shot evaluation.
Impact
Improves zero-shot repurposing predictions; clinical validation or adoption not stated in source.
Record
| Venue | Nature Medicine |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Kexin Huang, Payal Chandak, Qianwen Wang, Shreyas Havaldar, Akhil Vaid, Jure Leskovec, Girish N. Nadkarni, Benjamin Scott Glicksberg et al. |
| Citations | 239 (OpenAlex, at import time) |
| References | 63 |
| Model or method | TxGNN (graph neural network foundation model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41591-024-03233-x |
| Open access | hybrid |
| Keywords | drug repurposing, graph neural network, foundation model, zero-shot, knowledge graph |
Importance rating
significant — thousands to millions affected
Addresses diseases with limited or no treatment options across 17,080 diseases; predictions align with clinician off-label prescriptions in a large healthcare system.