DD 2.7: BBB Permeability Predictor
Fine-tuned ChemBERTa-77M-MLM on the BBBP dataset for binary classification of blood-brain barrier (BBB) permeability (permeable vs non-permeable).
Key Results (Scaffold Split Test Set, 13 Epochs)
- Accuracy: 84.48%
- F1 Score: 90.38%
- ROC-AUC: 0.8321
- PR-AUC: 0.9413
Training was done manually with PyTorch on a MacBook Air M1 (CPU only, due to accelerate/MPS compatibility issues).
Live Interactive Demo
Try the model right now without any code:
Gradio Space → BBB Predictor Demo
Features:
- Enter any SMILES string
- Get prediction + permeable probability
- See molecule visualization
- Pre-loaded examples (Ethanol, Aspirin, Caffeine, Doxorubicin)
How to Use
from transformers import AutoTokenizer, AutoModelForSequenceClassification
model_id = "Yousuf7/ChemBERT-BBB-Permeability"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id, num_labels=2)
- Downloads last month
- 1