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

This project is licensed under the MIT License.