CYP450 Metabolism Predictor

A web application for predicting Cytochrome P450 (CYP450) enzyme interactions with chemical compounds using Graph Convolutional Neural Networks.

Overview

This Streamlit-based application predicts whether chemical compounds are substrates or inhibitors of five major Cytochrome P450 enzymes. These enzymes are responsible for metabolizing approximately 70-80% of all drugs, making this tool valuable for pharmaceutical research, drug development, and drug-drug interaction (DDI) analysis.

Features

• Multi-Enzyme Prediction: Analyzes compounds against 5 key CYP450 enzymes:

  • CYP1A2
  • CYP2C9
  • CYP2C19
  • CYP2D6
  • CYP3A4

• Dual Prediction Types:

  • Substrate activity prediction
  • Inhibitor activity prediction

• Drug-Drug Interaction Analysis: Compare two molecules side-by-side to identify potential metabolic interactions

• Molecular Visualization: Automatically generates 2D structure images from SMILES input

• Molecular Property Calculation: Displays key molecular descriptors including:

  • Molecular Formula
  • Molecular Weight
  • LogP (lipophilicity)
  • TPSA (Topological Polar Surface Area)

• Pre-loaded Examples: Includes example molecule pairs demonstrating known DDI scenarios

Technology Stack

Core Frameworks

• Streamlit 1.19.0 - Web application interface
• DeepChem - Machine learning library for drug discovery
• TensorFlow 2.12.0 - Deep learning framework
• Keras 2.12.0 - Neural network API

Chemistry & Molecular Processing

• RDKit 2022.9.5 - Cheminformatics toolkit for SMILES parsing and molecular visualization

Data Science & ML

• NumPy 1.23.5 - Numerical computing
• Pandas 1.5.3 - Data manipulation
• Scikit-learn 1.2.1 - Machine learning utilities

Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Setup

1. Clone the repository:

git clone <repository-url>
cd deepchem_streamlit

2. Install dependencies:

pip install -r requirements.txt

Usage

Starting the Application

Run the Streamlit app:

streamlit run app.py

The application will open in your default web browser at http://localhost:8501

Using the Application

1. Enter Molecular Structures: Input one or two molecules in SMILES format

   • SMILES (Simplified Molecular Input Line Entry System) is a text notation for chemical structures
   • Example: CC(=O)Oc1ccccc1C(=O)O (Aspirin)

2. Use Example Molecules: Click example buttons to load pre-configured molecule pairs:

   • Example 1: Fluconazole & Warfarin (demonstrates CYP2C9 interaction)
   • Example 2: Fluoxetine & Metoprolol (demonstrates CYP2D6 interaction)

3. View Predictions: After entering SMILES, predictions are generated for:

   • Substrate activity (whether the compound is metabolized by each enzyme)
   • Inhibitor activity (whether the compound blocks enzyme activity)

4. Analyze Interactions: When comparing two molecules, the app automatically detects potential DDI scenarios:

   • Inhibitor + Substrate: One molecule may reduce metabolism of the other
   • Both Inhibitors: Potential additive inhibition effects
   • Both Substrates: Potential competitive metabolism

5. Explore Molecular Properties: Expand the properties section to view calculated descriptors

Model Architecture

Graph Convolutional Neural Networks

The application uses 10 trained Graph Convolutional Neural Network (GCN) models:

• 5 Inhibitor prediction models (one per enzyme)
• 5 Substrate prediction models (one per enzyme)

Model Details

• Type: DeepChem GraphConvModel
• Task: Binary classification (Active/Inactive)
• Training Data: Structure and activity data of known substrates and inhibitors
• Framework: TensorFlow/Keras

Model Storage

Models are stored in the models/ directory with the following structure:

models/
├── model_1A2_inh/     # CYP1A2 Inhibitor
├── model_1A2_sub/     # CYP1A2 Substrate
├── model_2C19_inh/    # CYP2C19 Inhibitor
├── model_2C19_sub/    # CYP2C19 Substrate
├── model_2C9_inh/     # CYP2C9 Inhibitor
├── model_2C9_sub/     # CYP2C9 Substrate
├── model_2D6_inh/     # CYP2D6 Inhibitor
├── model_2D6_sub/     # CYP2D6 Substrate
├── model_3A4_inh/     # CYP3A4 Inhibitor
└── model_3A4_sub/     # CYP3A4 Substrate

Each model directory contains TensorFlow checkpoint files with trained weights.

Application Architecture

Key Components

Model Loading (app.py:62-75)

• Models are loaded once at startup using @st.cache_resource
• Efficient caching prevents redundant loading across sessions

Featurization (app.py:77-82)

• ConvMolFeaturizer converts SMILES strings to graph representations
• Cached for performance optimization

Prediction Pipeline (app.py:84-126)

• Validates SMILES input
• Featurizes molecular structures
• Runs predictions across all 10 models
• Results cached based on SMILES input using @st.cache_data

Interaction Analysis (app.py:275-312)

• Automated detection of DDI scenarios
• Color-coded warnings for potential interactions

Use Cases

This application is designed for:

• Pharmaceutical Research: Early-stage drug candidate screening
• Drug Development: ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) profiling
• Drug-Drug Interaction Studies: Identifying potential metabolic interactions between medications
• Medicinal Chemistry: Structure-activity relationship analysis
• Toxicology: Predicting metabolic pathways and potential drug interactions

Limitations and Disclaimer

Important: These predictions are based on computational models and should be used for informational purposes only. They are not a substitute for:

• Experimental validation
• Professional medical advice
• Regulatory approval processes
• Clinical decision-making

Always validate computational predictions with appropriate experimental methods.

Performance Optimizations

• Model Caching: Models loaded once and reused across sessions
• Prediction Caching: Results cached by SMILES input to avoid redundant calculations
• Batch Processing: Predictions run in batches for efficiency
• Lazy Loading: UI components rendered on-demand

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

License

This project is licensed under the MIT License - see the LICENSE file for details.

References

CYP450 Enzymes

Cytochrome P450 enzymes are a family of enzymes involved in drug metabolism. The five enzymes covered in this application are responsible for metabolizing the majority of pharmaceutical compounds:

• CYP3A4: ~50% of drugs
• CYP2D6: ~25% of drugs
• CYP2C9: ~10% of drugs
• CYP2C19: ~10% of drugs
• CYP1A2: ~5% of drugs

Technologies

• DeepChem - Deep learning for drug discovery
• RDKit - Open-source cheminformatics
• Streamlit - Python web app framework
• TensorFlow - Machine learning platform

Support

For questions, issues, or feature requests, please open an issue in the repository.

---

Built with DeepChem, RDKit, and Streamlit