PathMoG is a graph neural network framework for multi-omics data integration and survival analysis, leveraging pathway knowledge to improve prediction accuracy and interpretability.
gnn/
├── moghet/ # Core model implementation
│ ├── core/ # Data loading and utilities
│ ├── data/ # Raw and processed data (excluded from GitHub)
│ ├── data_processing/ # Data processing scripts
│ ├── evaluation/ # Evaluation metrics
│ ├── inference/ # Inference utilities
│ ├── models/ # Model definitions
│ └── training/ # Training scripts
├── .gitignore # Git ignore rules
└── README.md # Project documentation
- Pathway-centric modular design: Leverages KEGG pathway knowledge to structure the graph architecture
- Multi-omics integration: Supports gene expression, mutations, and CNV data
- Heterogeneous Graph Transformer (HGT): Processes different node and edge types
- Survival prediction: Built-in Cox proportional hazards model for survival analysis
- Interpretable results: Identifies key genes and pathways associated with survival
# Create and activate environment
conda create -n pathmog python=3.9
conda activate pathmog
# Install dependencies
pip install torch torch-geometric
pip install pandas numpy scipy scikit-learn matplotlib seaborn lifelinesThe raw data is stored in moghet/data/raw/ and includes multi-omics data for various cancer types (BRCA, LUAD, GBM, etc.).
# Train the model on BRCA dataset
cd moghet
python training/train_hierarchical.py --dataset BRCA --epochs 100# Make predictions using a trained model
cd moghet
python inference/inference.py --model_path models/brca_model.pth --dataset BRCAmoghet/core/data_loader.py: Data loading and preprocessingmoghet/models/hierarchical_model.py: Main model definitionmoghet/training/train_hierarchical.py: Training scriptmoghet/inference/inference.py: Inference scriptmoghet/data_processing/build_hetero_graph.py: Heterogeneous graph constructionmoghet/evaluation/evaluate_trained_model_auc.py: Model evaluation
MIT License
For questions or collaboration, please contact the project maintainers.