pathwaygat 1.0.1

PathwayGAT is a GAT-based framework that could integrate information from different biological modalities, such as genes, microbes, and genetic variations. PathwayGAT organizes different elements into pathways and train an explainable GAT network to perform different tasks and reveal the underlying biological mechanism.

PathwayGAT

PathwayGAT is a GAT-based framework that could integrate information from genes, microbes, and genetic variations. Users can utilize the script provided in this repository to train their own model and obtain insights to the biological process they are interested in with the explanation value generated by the model.

Prerequisites

PathwayGAT requires the following Python package:

• NumPy
• pandas
• Matplotlib
• PyTorch
• PyTorch Geometric (https://pytorch-geometric.readthedocs.io/en/latest/)
• scikit-learn

Installation

It is recommended to create a separate conda environment for PathwayGAT. You can use the following command to install PathwayGAT from PyPI:

conda create -n test_pathwaygat python=3.9
conda activate test_pathwaygat
conda install numpy=1.26.4 pandas=2.2.2 matplotlib=3.8.4 torch-geometric scikit-learn
pip install pathwaygat

Run PathwayGAT

After installation, you can run the program with the following command:

pathwaygat <module> <parameters>

There are three kinds of modules for PathwayGAT.

• Training module: gene/microbe/microbe_gene/microbe_SNP/gene_SNP/microbe_gene_SNP/modules (run different kinds of GAT training depending on the input dataset availability)
• Evaluation module: evaluation (run cross-validation on the model parameters and plot ROC/PR curves)
• Explanation module: explanation (run GNNExplainer to explain the model, outputting the importance measurement of each feature in each pathway and sample)

You can also run the command pathwaygat <module> -h to display help information for each module, listing all the required and optional parameters for the framework.

Output

After running PathwayGAT training, evaluation, and explanation, you will obtain a list of output files:

• <output_prefix>.best_model.pth: The best PathwayGAT model obtained during training based on validation loss. It could be the model after the last epoch, but not necessarily.
• <output_prefix>.final_model.pth: The PathwayGAT model obtained after training for all the epochs.
• <output_prefix>_AUC.png: The ROC curve for the cross-validation step.
• <output_prefix>_AUPR.png: The PR curve for the cross-validation step.
• <output_prefix>_explanation_GNNExplainer_split.pt: The explanation output from GNNExplainer. The result is a list of Explanation objects from PyTorch Geometric. Each Explanation object contains explanation value for each feature in each node as node_mask in the shape of (node, feature).

Reference & Contact

If you have any questions about this package, please reach out to [e-mail].