Environmental setup

conda env create -f requirements.yaml

STpGCN for brain decoding

From the project's root folder, run

python main.py

to train the STpGCN or STpGCN-alpha or STpGCN-beta or STpGCN-gamma or STGCN or GAT or GCN or GIN or MLP-Mixer on the HCP S1200 datasets.

It would help if you manually changed the model_name to ensure you train the desired model.

NeurocircuitX for explainability

Run python NeurocircuitX_sep.py to calculate the importance score under keeping and masking strategy for each task.

After finishing the calculation of the importance score, you can run python NeurocircuitX_mix.py to calculate the final importance score for each task.

Main functions

• main.py:: The script serves as the entry point for the fMRI-based brain decoding pipeline. It orchestrates the entire process, from data loading and preprocessing to model training and evaluation.

• stpgcn_variants.py: The following spatial-temporal model variants are implemented:

  1. STGCN: The basic STGCN model with a configurable structure using the control_str parameter.
  2. STpGCN: STGCN with a pyramid structure for multi-scale temporal modelling.
  3. STGCN_hidden_feature: STGCN variant that outputs hidden features instead of final predictions.
  4. STpGCN_ab_bottom_up: STpGCN variant without a bottom-up pathway.
  5. STpGCN_ab_top: STpGCN variant without top pathway.
  6. ...

• layers.py:: This file contains the implementation of neural network layers used in stpgcn_variants.py.