Molecular Graph Property Prediction Project

This project implements a machine learning pipeline for predicting molecular properties from molecular graph representations. It utilizes RDKit for molecular processing, PyTorch Geometric for graph neural networks, and various machine learning techniques for prediction.

Table of Contents

• Project Overview
• Features
• Installation
• Usage
• Project Structure
• Configuration
• Dependencies
• Results
• Contributing
• License
• Author

Project Overview

This project aims to develop a robust and accurate model for predicting molecular properties based on their graph representations. It includes:

• Data preprocessing using RDKit for handling molecular data.
• Graph neural network models built with PyTorch Geometric.
• Training and validation pipelines with early stopping and learning rate scheduling.
• Evaluation metrics and visualization of results.

Features

• Modular Design: Code is organized into separate modules for better maintainability and readability.
• Configurable Pipeline: All parameters and processing steps are configurable via a YAML file.
• RDKit Integration: Utilizes RDKit for molecular data processing and feature extraction.
• PyTorch Geometric Models: Implements graph neural networks for molecular graph processing.
• Training Utilities: Includes early stopping, learning rate scheduling, and performance metrics.
• Visualization: Provides tools for plotting training/validation losses and performance metrics.

Installation

1. Clone the repository:

   git clone [https://github.com/yourusername/molecular-graph-prediction.git](https://www.google.com/search?q=https://github.com/yourusername/molecular-graph-prediction.git)
   cd molecular-graph-prediction

2. Create a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Linux/macOS
   venv\Scripts\activate  # On Windows

3. Install dependencies:

   pip install -r requirements.txt

Usage

1. Configure the project:

   • Modify the config.yaml file to set your desired parameters and data paths.

2. Run the main script:

   python main.py

3. View results:

   • The script will output training and validation metrics.
   • Plots of losses and metrics will be displayed.
   • Test set predictions and targets are saved as .npy files.

Project Structure

molecular-graph-prediction/ ├── config.yaml # Configuration file ├── main.py # Main script ├── dataset.py # Dataset loading and processing ├── data_utils.py # Data utility functions ├── models.py # Graph neural network models ├── rdkit_utils.py # RDKit utility functions ├── training_utils.py # Training and evaluation utilities ├── requirements.txt # Project dependencies └── README.md # Project documentation

Configuration

The config.yaml file allows you to configure various aspects of the project, including:

• Data paths and splits.
• RDKit processing steps.
• Model architecture and hyperparameters.
• Training parameters (learning rate, batch size, etc.).

Dependencies

• Python 3.x
• PyTorch
• PyTorch Geometric
• RDKit
• scikit-learn
• numpy
• matplotlib
• PyYAML
• pydantic

Install these dependencies using pip install -r requirements.txt.

Results

The project outputs:

• Training and validation loss curves.
• Performance metrics (MAE, MSE, R2, Explained Variance) on the validation and test sets.
• Saved .npy files containing test set targets and predictions.

Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository.
2. Create a new branch for your feature or bug fix.
3. Make your changes and commit them.
4. Push your changes to your fork.
5. Submit a pull request.

License

This project is licensed under the MIT License.

Author

Asadollah Boshra/shahram-boshra - [email protected] / https://github.com/shahram-boshra