FDGNN (Fingerprints and Descriptors with Graph Neural Networks)

Project Description

FDGNN is a research-focused project for Molecular Property Prediction using Graph Neural Networks (GNNs). It integrates molecular fingerprints and descriptors with GNNs for comprehensive molecular data modeling. The project includes data preprocessing, feature extraction, model training, evaluation, and visualization.

Table of Contents

• Project Description
• Folders Structure
• Installation
• Usage
• Modules
• Notebooks
• Contributing
• License

Folders Structure

Thesis/
│
├── data/
│   ├── ...
│   ├── README.md
│   └── datasets/
│       ├── bace.csv
│       ├── BBBP.csv
│       ├── clintox.csv
│       ├── Esol.csv
│       ├── freesolv.csv
│       ├── HIV.csv
│       ├── Lipophilicity.csv
│       └── ...
│   # Contains data used for models
│
├── examples/
│    └── ...
│   # Contains various examples of model training in different scenarios (different datasets, Random split, scaffold split, Cross-validation, etc.)
│
├── models/
│    └── GnnPooling.py
│   # Proposed model code and structure
│
├── modules/
│   ├── __init__.py
│   ├── data_handler.py
│   ├── fingerprints_descriptors.py
│   ├── utils_classification.py
│   ├── utils_cls_cv.py
│   ├── utils_reg_cv.py
│   └── utils_regression.py
│   # Various functions and classes for data preparation, dataset construction in PyG, model training, and evaluation
│
├── notebooks/
│   └── ...
│   # Notebooks for testing modules
│
├── test_results/
│   ├── Bace.ipynb
│   ├── Bace2.ipynb
│   ├── BBBP.ipynb
│   ├── Esol.ipynb
│   ├── Freesolv.ipynb
│   └── Lipo.ipynb
│   # **Contains final model results on datasets**
│
├── .gitignore
├── README.md
└── requirements.txt

Installation

To set up the project locally, follow these steps:

Prerequisites

Ensure you have Python installed (recommended: Python 3.8+). Then install the required dependencies:

pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.3.0+cu121.html
pip install torch_geometric
pip install deepchem
pip install rdkit
pip install -q ogb
pip install dgllife
pip install torchinfo
pip install molfeat

Usage

Running a Notebook

To explore datasets and train models interactively, navigate to the notebooks/ directory and run:

jupyter notebook notebooks/data_handler.ipynb

Modules

The modules/ directory contains the core functionalities:

• data_handler.py: Functions and classes for data preprocessing.
• fingerprints_descriptors.py: Molecular fingerprint and descriptor calculations.
• utils_classification.py: Utility functions for classification tasks.
• utils_regression.py: Utility functions for regression tasks.
• utils_cls_cv.py & utils_reg_cv.py: Cross-validation utilities for classification and regression.

Contributing

We welcome contributions! To contribute:

1. Fork the repository.
2. Create a new branch (git checkout -b feature-branch).
3. Commit your changes (git commit -m 'Add new feature').
4. Push to your branch (git push origin feature-branch).
5. Open a Pull Request.

License

This project is licensed under the MIT License. See LICENSE for details.