This repository consists of scripts for reproducing results in the "Modelling variability in dynamic functional brain networks using embeddings" manuscript.
- Scripts for preprocessing data depends on the osl toolbox.
- Scripts for training models and analysing results depends on the osl-dynamics toolbox, which includes source code for the HIVE model, as well as analysis tools.
data_preprocessing: This directory contains scripts for preprocessing, coregistration, source reconstruction and fixing sign ambiguity for the three MEG datasets used.
simulations: This directory contains scripts for simulation analysis on HIVE.
simulation_1.py: This script shows how covariance deviations is learnt by the variability encoding block in HIVE.simulation_2.py: This script shows how the underlying subpopulation structure is inferred by HIVE.simulation_3.py: This script shows HIVE performs more accurate inference than HMM-DE and can make use of increasing amount of heterogeneous data.
real_data: This directory contains scripts for training HIVE and HMM-DE, and perform analysis on three real MEG datasets.
wakeman_henson: This directory contains scripts for training, analysing HIVE and HMM-DE on the Wakeman-Henson dataset.multi_dataset: This directory contains scripts for training, analysing HIVE and HMM-DE on combined resting-state data from the MRC MEGUK Nottingham site dataset and the Cam-CAN dataset.camcan: This directory contains scripts for training HIVE and HMM-DE on the Cam-CAN dataset resting-state data. There is also a script for performing age prediction from inferred features from both approaches.