| title | StereoAwareGNN BBB Predictor |
|---|---|
| emoji | 🧠 |
| colorFrom | green |
| colorTo | blue |
| sdk | docker |
| app_file | app.py |
| pinned | false |
A graph neural network for blood-brain-barrier permeability prediction.
Correction (2026). An earlier version of this README advertised "0.9612 AUC, state-of-the-art on external validation." That figure came from training on BBBP and testing on B3DB, which overlaps BBBP by ~22%, so it was not true external validation. Under leakage-controlled evaluation (scaffold splitting) the model scores roughly 0.83-0.84 AUC, is competitive with but does not clearly beat a simple ECFP + random-forest baseline, and its stereo features do not affect BBB predictions (it returns identical outputs for enantiomers). A full, reproducible audit is here: https://github.com/abinittio/bbb-honest-eval
Nabil Yasini-Ardekani
- Graph neural network for BBB permeability prediction
- Real-time BBB permeability prediction
- Molecular visualization
- Export results as JSON/CSV