Multi-Omics Deep Learning for Disease Prediction

This work is part of an IEEE research paper submission and is intended solely for academic research purposes. Use or redistribution without proper permission is prohibited.

This repository integrates multi-omics data (genomics, transcriptomics, proteomics, and metabolomics) with deep learning models to predict autoimmune and neurodegenerative diseases. By combining statistical and representation learning with explainable AI (XAI) methods, the project not only delivers high predictive accuracy but also provides insights into the biological drivers behind each prediction.

Theoretical Framework

• Deep Learning Models

  • Feedforward Neural Networks (FNN): Extract static omics features to build high-level representations.
  • Recurrent Neural Networks (RNN): Capture temporal dependencies inherent in biological processes.
  • Bidirectional LSTM & GRU (BiLSTM, BiGRU): Analyze bidirectional dependencies to better model complex biological pathways.

• Explainable AI (XAI) Methods

  • SHAP: Utilizes game theory to assign each omics feature a contribution value, offering both global and local interpretability.
  • LIME: Builds locally interpretable surrogate models that explain individual predictions.
  • Evaluation on 4 Diseases: XAI techniques are applied across four distinct diseases, enabling comparative analysis of key biomarkers and ensuring that model decisions are transparent and reliable across diverse clinical contexts.

Implementation

1. Repository Setup

   • Clone the repository and navigate to the project directory.

2. Dependency Installation

   • Ensure Python 3.7+ is installed.
   • Install required libraries:

     pip install tensorflow keras shap lime

3. Data Processing

   • Place the omics datasets (sourced from public repositories like GEO and TCGA) in the designated folder.
   • Run the preprocessing scripts to clean and structure the data.

4. Model Training & Evaluation

   • Train deep learning models using the preprocessed multi-omics data.
   • Evaluate model performance using standard metrics.
   • Utilize SHAP and LIME to interpret model decisions and validate biomarker relevance across four different diseases.

Contributing

Contributions to this repo are welcome. Please open an issue or submit a pull request if you have suggestions for:

• New feature extraction and data preprocessing techniques.
• Additional machine learning or deep learning models.
• Enhanced visualization and interpretability tools.

Before using or modifying the project, please note that it is part of an IEEE research paper submission and should not be used without explicit permission.

License

This project is licensed under the MIT License.

Contact

For questions, collaborations, or further details, please reach out:

• Name: Shib Kumar
• Email: [email protected]
• GitHub: @shib1111111