README.md

metadata

title: Multi-Endpoint Toxicity Predictor
colorFrom: blue
colorTo: purple
sdk: streamlit
sdk_version: 1.28.0
app_file: app.py
pinned: false
license: mit

Multi-Endpoint Toxicity Predictor

Predict drug toxicity across three complementary cellular mechanisms using machine learning.

What does this predict?

This tool predicts toxicity across three complementary endpoints:

1. ** Oxidative Stress (ARE/Nrf2)** - ROC-AUC: 0.82

   • Cellular antioxidant response
   • ROS-mediated damage

2. ** Mitochondrial Dysfunction (MMP)** - ROC-AUC: 0.93

   • Membrane potential disruption
   • Energy depletion

3. ** DNA Damage (p53)** - ROC-AUC: 0.82

   • Genotoxic stress
   • Mutagenic potential

How to use

1. Enter a SMILES string (e.g., CC(=O)Oc1ccccc1C(=O)O for Aspirin)
2. Click Predict
3. View results across all three endpoints
4. Get molecular property analysis and design suggestions

Model Details

Training Data:

• 11,306 compounds from EPA ToxCast database
• 2,071 molecular features (23 descriptors + 2,048 fingerprints)
• Random Forest with class balancing

Performance:

• 5-fold cross-validation
• External validation: 100% accuracy on known toxic/safe compounds
• Calibrated thresholds for optimal sensitivity

Risk Levels:

• 🟢 LOW (<35%): Safe profile
• 🟡 MODERATE (35-60%): Requires evaluation
• 🔴 HIGH (>60%): Significant concerns

The Science

Our analysis revealed that toxicity follows a predictable cascade:

Lipophilic Compound → Mitochondrial Accumulation
                   ↓
        Membrane Disruption (MMP)
                   ↓
           ROS Production
                   ↓
      Oxidative Stress (ARE)
                   ↓
       DNA Damage (p53)

Key Findings:

• LogP 4-6 compounds: 10x higher toxicity
• ≥3 aromatic rings: 3.5x higher mitochondrial toxicity
• MW >400 Da: 2.4x higher DNA damage
• 57% of MMP+ compounds also activate ARE

Disclaimer

For research purposes only. Predictions should be validated experimentally before making clinical decisions. This tool is not approved for regulatory submissions.

Citation

If you use this tool in your research, please cite:

Multi-Endpoint Toxicity Predictor: A Mechanistic Framework
EPA ToxCast Database (2024)
https://huggingface.co/spaces/MlchaeI/Toxicity_2

Learn More

• Data Source: EPA ToxCast
• Trained on: 11,306 compounds
• Model: Random Forest ensemble

Contributing

Found a bug? Have suggestions? Please open an issue!

---

Built with love for the drug discovery community