Effective connectivity estimation

A practical guide to estimating directed connectivity from regional time series with the multivariate Ornstein–Uhlenbeck model.

Follow the pipeline, inspect the choice of decay timescale τ, and compare connectivity patterns across the convergent range.

Installation

Download or clone the source repository, then open its ec_estimation folder. It includes the local pyMOU source used in the analyses.

git clone https://github.com/benedetta-mariani/mOU-EC-estimation.git
cd mOU-EC-estimation/ec_estimation
python -m pip install -r requirements.txt

Use the bundled pyMOU-master folder. The pipeline does not require installing pyMOU from PyPI.

Quick start

1. Open EC_pipeline_precision.ipynb in your Jupyter environment.
2. Set the participant, input file, TR, mapping and τ grid in the first section. The input array has shape time × regions.
3. Run the cells in order to prepare the data, fit the models and select τ.
4. Inspect the selection curve, the selected connectivity matrix and the correlations across τ.

The example uses HCP regional time series and a 100-to-62 cortical mapping. Adapt the region selection and mapping to your own data.

Choosing τ

The pipeline selects an interior local minimum of the off-diagonal correlation with a precision-based reference, within the retained convergent interval. The notebook documents the exclusions and diagnostics. This is an operational selection criterion, rather than a uniquely identified physiological timescale.

Saved results

Only the selected matrices and compact diagnostics are saved. The full grids of EC and noise matrices stay in memory for the comparisons. Saved results can be reloaded without fitting the models again.

These pages display the saved code and figures. To execute the analysis, use the notebook in Python. Input datasets are not included.

Source & citation

The analysis repository contains the pipeline and its source code. This documentation displays the published source version.

Model estimation uses pyMOU, developed by Andrea Insabato, Gorka Zamora-López and Matthieu Gilson. The fitting method refers to Gilson et al. (2016; 2020). See the pyMOU reference for the implementation used here and its license.