GONet

GONet establishes a robust and generalizable graph representation learning for integrating multi-omics data (mRNA, miRNA, and proteomics) and interaction networks.

Requirements

Python >= 3.8

PyTorch >= 2.0

PyTorch Geometric >= 2.4.0

Scikit-learn, Pandas, Numpy, Matplotlib

Installation

git clone https://github.com/yourusername/GONet.git

cd GONet

conda env create -f environment.yml

conda activate GONet

Data Input

GONet expects input data in CSV format. Each omic layer should be a matrix of (samples × features). We provide a processed data in the data/ directory, and place your raw data in same path if you want to test your data.

Usage

GONet provides a streamlined pipeline from hyperparameter optimization to final model training.

1. Configuration

All model parameters and experiment settings are managed in args.py. Before running, ensure you have configured the paths and basic settings (e.g., learning rate, weight decay, or GNN layers) in this file.

2. Hyperparameter Optimization (Optional)

To find the optimal hyperparameters for a specific cancer dataset, we utilize Optuna. This step is recommended for achieving peak performance on new data.

python3 main_optuna_model.py

The search space and number of trials can be adjusted within main_optuna_model.py.

3. Model usage

You can train GONet using either the Python script or the provided shell script for batch processing.

Option A: Direct usage Specify the target cancer type using the --cancer_type argument:

python3 main_train.py --cancer_type HTML_THCA

Option B: Batch/Shell usage Use the shell script to execute training with predefined environment settings (recommended for server-side execution):

run_train_new.sh HTML_THCA

The model outputs performance metrics including AUC-ROC, F1-score, and Accuracy.