GCCN

source code of Graph Capsule Convolutional Network (GCCN).
The data used in this study can be openly available at http://adni.loni.usc.edu/.

Dependencies

The script has been tested running under Python 3.9.0, with the following packages installed (along with their dependencies):

• numpy==1.21.2
  
• networkx==2.6.3
  
• torch==1.11.0
  
• torch-geometric==2.0.4
  In addition, CUDA 11.3 have been used on NVIDIA GeForce RTX 3080.
  

Overview

The repository is organised as follows:

• dataset.py: contains the implementation of Heterogeneous Pathogenic Information Association Graphs (HPIAGs);
  
• transform.py: include the implementation of batching operation and graph related feature engineering;
  
• disentangle.py: contains a variety implementation of disentangling functions;
  
• denseconv.py: is an simply implementation of DenseGCNConv;
  
• layers.py: implements the three layers (primary layer & digital layer & reconstruction layer);
  
• models.py: contains the implementation of the HGCCN;
  
• custom_function.py: contains the HGCCN related operation;
  
• sparsemax.py: contains the implementation of Sparsemax;
  
• parameters.py: including all the parameters involoved in model;
  
• train_eval_helper.py: contains the cross-validation related helper functions.
  
• Finally, main.py puts all of the above together and be used to execute a full training run.