Task-specific model
Open weights
Research use only
drug-discovery
TxGNN
Graph neural network for zero-shot drug repurposing that scores drug–disease indications and contraindications over a medical knowledge graph of 17,080 diseases, including ones with no approved treatment.
At a glance
DeveloperZitnik Lab, Harvard Medical School
Released2024-09-25
LicenceMIT
AvailabilityOpen weights
KindTask-specific model
Regulatory statusResearch use only
What it does
Not yet documented on this card.
Tasks, data types and cancers
Data typeSmall moleculesBiomedical literature
Clinical settingDrug discovery
CancerPan-cancer
Inputknowledge graph (drugs, diseases, genes, phenotypes)
Outputindication / contraindication scores with explanatory paths
Architecture
FamilyGraph neural network with disease-similarity metric learning
Training data
PrimeKG-derived medical knowledge graph; evaluated on diseases held out entirely from training (zero-shot).
Evaluation
| Benchmark / dataset | Metric | Value | External validation | Source |
|---|---|---|---|---|
| zero-shot held-out diseases | AUPRC improvement over baselines | 49.2% (indications), 35.1% (contraindications), per paper | yes | Source |
How to run
LibraryPyTorch Geometric (repo)
Pretrained model and an explorer UI are linked from the repository.
Regulatory status and intended use
Regulatory statusResearch use only
Intended useHypothesis generation for repurposing research.
Regulatory status is quoted from the source linked above and can change. Research-use-only models must not be used for clinical decisions.
Limitations and bias
- Predictions are hypotheses over knowledge-graph edges; clinical evidence is required before any use.
- Oncology indications are a subset of a general-purpose graph.
Sources
This page is educational — it is not medical advice and does not replace consultation with an oncologist. Diagnostic and treatment decisions are made solely by specialist physicians.