Spaces:

nabilyasini
/

StereoAwareGNN1

Sleeping

App Files Files Community

StereoAwareGNN1 / QUICK_START.md

nabilyasini

Upload folder using huggingface_hub

84766d8 verified 4 months ago

preview code

raw

history blame contribute delete

6.73 kB

BBB Permeability Predictor - Quick Start Guide

Get started with BBB predictions in 3 easy steps!

🚀 Quick Start (3 Steps)

Step 1: Launch the Web Interface

Windows:

# Double-click this file
launch_web.bat

Command Line:

streamlit run app.py

Step 2: Select a Molecule

Choose from three input methods:

Common Molecules - Pick from 20+ pre-loaded drugs
SMILES String - Paste any SMILES notation
Molecule Name - Type the drug name (beta)

Step 3: Get Predictions!

Click "Predict BBB Permeability" and instantly see:

✅ BBB+ (High permeability)
⚠️ BBB± (Moderate permeability)
❌ BBB- (Low permeability)

📊 What You Get

Instant Results

BBB Permeability Score (0.0 - 1.0)
Category Classification (BBB+/BBB±/BBB-)
Confidence Level

Detailed Analysis

Molecular Properties
- Molecular Weight
- LogP (lipophilicity)
- TPSA (polar surface area)
- H-bond donors/acceptors
Drug-likeness Metrics
- Lipinski's Rule of 5
- BBB-specific rules
- Warnings for suboptimal properties

Beautiful Visualizations

📊 Gauge Chart - BBB score meter
🕸️ Radar Chart - Drug-likeness profile
📈 Bar Chart - Property distribution

Export Options

💾 Download results as CSV
📄 Download results as JSON

🎯 Example Predictions

Example 1: Caffeine (CNS Drug)

Input: Caffeine (or SMILES: CN1C=NC2=C1C(=O)N(C(=O)N2C)C)
Output:
  BBB Score: 0.782
  Category: BBB+ ✅
  Interpretation: HIGH BBB permeability
  MW: 194.2 Da | LogP: -1.03 | TPSA: 61.8 A^2

Example 2: Glucose (Sugar)

Input: Glucose (or SMILES: C(C(C(C(C(C=O)O)O)O)O)O)
Output:
  BBB Score: 0.109
  Category: BBB- ❌
  Interpretation: LOW BBB permeability
  MW: 180.2 Da | LogP: -3.24 | TPSA: 110.4 A^2

Example 3: Benzene (Aromatic)

Input: Benzene (or SMILES: c1ccccc1)
Output:
  BBB Score: 0.802
  Category: BBB+ ✅
  Interpretation: HIGH BBB permeability
  MW: 78.1 Da | LogP: 1.69 | TPSA: 0.0 A^2

🔬 Pre-loaded Molecules

The app includes 20+ common molecules across 4 categories:

CNS Drugs (8 molecules)

Caffeine
Cocaine
Morphine
Nicotine
Aspirin
Ibuprofen
Acetaminophen
Propranolol

Simple Molecules (4 molecules)

Ethanol
Benzene
Toluene
Glucose

Amino Acids (3 molecules)

Glycine
Alanine
Tryptophan

Neurotransmitters (3 molecules)

Dopamine
Serotonin
GABA

💡 Tips for Best Results

Using SMILES Input

Get SMILES from databases like:
- PubChem
- ChEMBL
- DrugBank
Paste the SMILES string directly
Click "Predict BBB Permeability"

Understanding Results

BBB+ (Score ≥ 0.6)

✅ Likely crosses blood-brain barrier
✅ Potential CNS activity
✅ Good for neurological drugs

BBB± (Score 0.4-0.6)

⚠️ Moderate permeability
⚠️ Case-by-case evaluation needed
⚠️ May require optimization

BBB- (Score < 0.4)

❌ Unlikely to cross BBB
❌ Peripheral action only
❌ Not suitable for CNS targets

Interpreting Warnings

Common warnings and what they mean:

"High molecular weight (>450 Da)"

Large molecules struggle to cross BBB
Consider reducing molecular size

"LogP outside optimal range (1-5)"

Too hydrophilic (LogP < 1): Poor membrane penetration
Too lipophilic (LogP > 5): Poor solubility

"High TPSA (>90 A^2)"

Too polar to cross BBB efficiently
Reduce polar surface area

"High H-bond donors (>3)"

Too many H-bond donors reduce permeability
Mask or remove donor groups

🛠️ Troubleshooting

Problem: "Model not found"

Solution: Train the model first

python train_gnn.py

Problem: "OpenMP Error"

Solution: Set environment variable

set KMP_DUPLICATE_LIB_OK=TRUE  # Windows
export KMP_DUPLICATE_LIB_OK=TRUE  # Linux/Mac

Problem: Web interface won't start

Solution: Install dependencies

pip install streamlit plotly

Problem: Port already in use

Solution: Use different port

streamlit run app.py --server.port 8502

📚 Additional Resources

Documentation

README.md - Complete system documentation
WEB_INTERFACE.md - Web UI details
RESULTS.md - Performance metrics

Code Examples

app.py - Web interface code
predict_bbb.py - Prediction API
demo.py - Command-line examples
train_gnn.py - Training pipeline

Research Background

BBB permeability is critical for CNS drug development
Only ~2% of small molecules cross the BBB
Our GNN model achieves MAE of 0.0967 on validation set

🎓 Understanding BBB Permeability

What is the Blood-Brain Barrier?

The BBB is a selective barrier that protects the brain from harmful substances while allowing nutrients to pass through.

Why is it Important?

Drug Development: CNS drugs must cross BBB
Toxicity: Non-CNS drugs should NOT cross BBB
Neurological Diseases: BBB permeability affects treatment efficacy

Key Factors for BBB Crossing

Small Size (MW < 450 Da)
Moderate Lipophilicity (LogP 1-5)
Low Polarity (TPSA < 90 Ų)
Few H-bond Donors (≤3)
Few H-bond Acceptors (≤7)

🌟 Key Features

Model Specifications

Architecture: Hybrid GAT+GraphSAGE
Parameters: 649,345
Validation MAE: 0.0967
Training Dataset: 42 curated compounds
Prediction Time: <1 second

Web Interface Features

✨ Modern gradient UI design
📱 Responsive layout
🎨 Interactive visualizations
💾 Export to CSV/JSON
🔍 Real-time predictions
📊 Comprehensive analysis
⚠️ Intelligent warning system

🚀 Next Steps

Try the Web Interface
```
launch_web.bat
```
Test Some Molecules
- Start with pre-loaded molecules
- Try your own SMILES strings
Analyze Results
- Compare BBB+ vs BBB- molecules
- Understand property distributions
Export and Share
- Download results as CSV
- Share predictions with team
Explore Advanced Features
- Read WEB_INTERFACE.md
- Check README.md
- Run python demo.py for API examples

📞 Support

For questions or issues:

Check this Quick Start guide
Review WEB_INTERFACE.md
See README.md for technical details
Run python demo.py for usage examples

Ready to predict BBB permeability?

# Launch the web interface now!
streamlit run app.py

Enjoy using the BBB Permeability Predictor! 🧬✨