This repository contains the necessary code and a minimal working example of preprocessed data to reproduce the results of:

" The structure and statistics of language jointly shape cross-frequency neural dynamics during spoken language comprehension", H. Weissbart & AE. Martin, 2024 (Nature Communications).

• Content
• Setting up
• Dependencies
• Data
• Reference

Folders content:

• data
  • preprocessed MEG file: sample files are provided in as a figshare dataset (see data/meg/README.md for more information)
  • Stimulus:
    • Audio files: data/stim/ contains the audio files used in the experiment
    • Annotations: All annotations are stored in the data/stim/annotations folder
• audiobook: a custom made python package containing some utility functions specific to our "audiobook MEG" dataset (see the dataset section for more information)
• pyeeg: a custom made python package containing some utility functions for computing TRFs (listed as submodule here)
• notebooks: example Jupyter Notebook reproducing some of the main results and figures of the paper

The overall installation process is simple and fast, given that the main dependencies are already pre-installed in the provided conda environment. Installing from scratch in a new conda environment should take only a few minutes. It is mainly about setting up the right path for some of the utility functions.

Firstly, clone the repository and navigate to the root folder of the repository:

git clone https://github.com/Hugo-W/feature-PAC
cd feature-PAC 

The main requirement is the mne library. You can install it using with conda:

conda install -c conda-forge mne

More information on how to install mne can be found here.

Following the installation of mne, you must install the custom package pyeeg. This package contains utility functions for computing TRFs and is included as a submodule in this repository. To install it, run the following commands from the root directory (making sure you have the correct conda environment activated):

git submodule update --init
cd pyeeg
pip install .
cd ..

The present code depends on python modules within the audiobook folder. This is not packed as a package, so you must add the path to the audiobook folder to your PYTHONPATH. The easiest is to run code directly within a directory containing audiobook folder. Alternatively, you can add it to the path by running the following command from the root directory:

export PYTHONPATH=$PYTHONPATH:$(pwd)/audiobook

Or from within python (either in a script or in a Jupyter notebook) you can add the path to the audiobook folder as follows:

import sys
sys.path.append('path/to/feature-PAC/audiobook')

Example code to reproduce the results of the paper is provided in the notebooks folder. You can run the notebooks directly from the command line or from a Jupyter notebook server (e.g., Jupyter Lab jupyter lab ./ from the root directory, then navigate to the notebook folder within Jupyter).

The code was tested with Python 3.12.2 and 3.9.13 on CentoOS 7 (kernel Linux-5.4.230-1.el7.elrepo.x86_64-x86_64-with-glibc2.17) and Windows 10 respectively with the following dependencies (mostly scientific python libraries):

• mne 1.6.1
• numpy 1.26.4
• scipy 1.13.0
• matplotlib 3.8.4
• sklearn 1.4.2
• pandas 2.2.2
• joblib 1.4.0
• seaborn 0.13.2
• statsmodels 0.14.1
• h5py >=3.6.0
• h5io >=0.1.0
• fooof >=1.0.0

A minimal working example of the preprocessed data is provided as a figshare dataset. The data is stored in a MNE Raw object, which can be loaded using the mne.io.read_raw_fif function. The data is stored in a file named sub037-audioBook-filtered-ICAed-raw.fif. The file contains the MEG data, the channel names, the sampling frequency. Events are stored in audioBook-eve.fif, and a covariance matrix is stored in story-cov.fif. These files allow for a single subject analysis.

The raw files of the full dataset used in this study are published and available in the Donders repository:

Martin, A.E. (2023): Constructing sentence-level meaning: an MEG study of naturalistic language comprehension. Version 1. Radboud University. (dataset). https://doi.org/10.34973/a65x-p009

Paper:

The Structure and Statistics of Language jointly shape Cross-frequency Neural Dynamics during Spoken Language Comprehension; Hugo Weissbart, Andrea E. Martin; Nature Comms. (2024);

Code:

Weissbart, H. (2024). Code and example for TRF-PAC (v0.1.0). Zenodo.