Deep Learning for Molecular and Genomic Characterization of Lung Cancer in Never-Smokers Using Hematoxylin and Eosin-Stained Images

Mutation_AI is a deep learning project for mutation analysis using Convolutional Neural Networks (CNNs) with two main modules:

• Multilabel_CNN: For multi-label classification tasks.
• Binary_CNN: For binary classification tasks.

Both modules implement a custom ResNet-50-like architecture using TensorFlow/Keras.

---

---

Table of Contents

• Installation
• Project Structure
• Usage
  • Multilabel CNN
  • Binary CNN
• Data Structure
• Model Architecture
• Requirements
• Contributing
• License

---

Installation

1. Clone the repository

   git clone https://github.com/monjoybme/Mutation_AI.git
   cd Mutation_AI

2. Install dependencies

   It is recommended to use a Python virtual environment.

   python3 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   pip install -r requirements.txt

---

Project Structure

Mutation_AI/
│
├── Multilabel_CNN/
│   ├── model.py         # Custom ResNet-50 model for multilabel classification
│   ├── main.py          # Training/Inference script (if available)
│   └── ...              # Additional utilities and scripts
│
├── Binary_CNN/
│   ├── model.py         # Custom ResNet-50 model for binary classification
│   ├── main.py          # Training/Inference script (if available)
│   └── ...              # Additional utilities and scripts
│
├── requirements.txt
└── README.md

---

Usage

Multilabel CNN

1. Prepare your dataset as described in the Data Structure section.
2. Configure parameters in Multilabel_CNN/main.py (if available).
3. Run the training script:

   python Multilabel_CNN/main.py

Binary CNN

1. Prepare your dataset as described in the Data Structure section.
2. Configure parameters in Binary_CNN/main.py (if available).
3. Run the training script:

   python Binary_CNN/main.py

---

Data Structure

• The models expect input data in a format compatible with TensorFlow/Keras.

• Typical directory structure for image data:

  data/
    train/
      class_1/
        img001.png
        img002.png
        ...
      class_2/
        ...
    val/
      class_1/
      class_2/
  

• For multilabel tasks, a CSV file with file paths and corresponding label vectors is often used.

• For binary tasks, two folders (e.g., positive/ and negative/) or a similar structure.

Note: Adjust data loading utilities as needed for your specific data organization.

---

Model Architecture

Both Multilabel_CNN/model.py and Binary_CNN/model.py implement a custom ResNet-50-like architecture using TensorFlow/Keras:

• Initial convolutional and pooling layers
• Multiple custom residual blocks (with optional shortcuts)
• Global Average Pooling
• Dense layers with dropout
• Output layer:
  • sigmoid activation for multilabel and binary classification

The output layer's size is determined by the number of classes (multi-label) or 1 (binary).

---

Requirements

See requirements.txt for full details.

Main dependencies:

• tensorflow>=2.0
• numpy
• pandas
• scikit-learn
• matplotlib (optional, for plotting)
• tqdm (optional, for progress bars)

---

Contributing

Contributions are welcome! Please open issues or pull requests for improvements or bug fixes.

---

License

© Monjoy.AI

This code is made available under the GNU General Public License v3.0 (GPLv3).

You are free to use, modify, and redistribute this software in accordance with the terms of the GPLv3 license.

This software is provided for non-commercial academic purposes. Any commercial use, distribution, sublicensing, or incorporation into proprietary products requires prior written permission from the copyright holder.

See the LICENSE file for the full GPLv3 license text.

© Monjoy.AI. All rights reserved.