Strock_Nghiem_bioRxiv_2025

This project explores how feature attribution methods can identify the source of excitation/inhibition (E/I) imbalances in simulated fMRI data.

Setting up environment

For TVB simulations, use requirements_tvb.txt, for classification training and feature attribution, use requirements_torch.txt and prior to running scripts run:

source environment.sh

RNN simulation

Training all the models

bash model/rnn_exc/train_all_parameter.sh

Test all the models and computing feature attribution

bash model/rnn_exc/test_all_parameter.sh
bash model/rnn_exc/test_all_parameter_baseline.sh

TVB human simulation

Generate simulated data

python dataset/tvb/EXAMPLE_human.py
python dataset/tvb/EXAMPLE_mouse.py

Training all the models

bash model/tvb/train_all_parameter_human.sh
bash model/tvb/train_all_parameter_mouse.sh

Test all the models and computing feature attribution

bash model/tvb/test_all_parameter_human.sh
bash model/tvb/test_all_parameter_baseline_human.sh
bash model/tvb/test_all_parameter_mouse.sh
bash model/tvb/test_all_parameter_baseline_mouse.sh