ecomp_analysis

This repository contains the analysis code for the eComp project.

The archived version can be found on Zenodo:

• doi: 10.5281/zenodo.6411287

All important details are reported in the original paper for the project:

• preprint: 10.1101/2022.03.31.486560
• journal article: 10.1371/journal.pcbi.1010747

Installation

To install a Python environment suitable to running this analysis code, please:

1. Download Miniconda for your system: https://docs.conda.io/en/latest/miniconda.html (this will provide you with the conda command)
2. Use conda to install mamba: conda install mamba -n base -c conda-forge (for more information, see: https://github.com/mamba-org/mamba)
3. Use the environment.yml file in this repository to create the ecomp_analysis environment: mamba env create -f environment.yml

You can now run conda activate ecomp_analysis and should have all required packages installed for running the analyses.

Obtaining the data

The code can run with the mpib_ecomp_sourcedata dataset:

• repository: https://gin.g-node.org/sappelhoff/mpibecompsourcedata
• doi: 10.12751/g-node.lir3qw

However, running the code the first time might take a long time in order to produce all derivatives that are not shipped as part of this (analysis code) repository. Downloading the mpib_ecomp_derivatives dataset may speed this step up (not required):

• repository: https://gin.g-node.org/sappelhoff/mpibecompderivatives
• doi: 10.12751/g-node.9rtg6f

There is a Brain Imaging Data Structure (BIDS) version of the data available:

• repository: https://gin.g-node.org/sappelhoff/mpibecompdataset
• doi: 10.12751/g-node.jtfg5d

Finally, there is the experiment code that was used when the data was collected:

• repository: https://github.com/sappelhoff/ecomp_experiment/
• doi: 10.5281/zenodo.6411313

Configuration before running the code

Before running the code on your machine, make sure you have activated the ecomp_analysis environment, and have downloaded all data (see steps above).

Next, you need to open the config.py file in this repository. In there, you will find several important settings that can be kept stable. However, you will need to adjust the path variables pointing to the data, for example:

• DATA_DIR_LOCAL, pointing to the data stored locally
• DATA_DIR_EXTERNAL, pointing to the data stored on an external harddrive
• ANALYSIS_DIR_LOCAL, pointing to the analysis code repository (THIS repository) stored locally
• ...

Generally, the analyses scripts will import the paths defined in config.py and use them to obtain the data. For example:

```Python from config import DATADIREXTERNAL, ANALYSISDIRLOCAL

datadir = DATADIREXTERNAL analysisdir = ANALYSISDIRLOCAL

... use datadir and analysisdir

```

you can edit the analysis scripts themselves to get the data from local, external, or remote sources that you have specified in config.py.

Running the code

Some of the analysis scripts (specifically those starting with two digits, e.g., 01_find_bads.py) accept command line options. It is convenient to run them from the command line (e.g., with bash) like so:

bash for i in {1..32} do python -u 01_find_bads.py \ --sub=$i \ --data_dir="/mnt/home/appelhoff/Projects/ARC-Studies/eeg_compression/ecomp_data/" \ --analysis_dir="/mnt/home/appelhoff/Projects/ARC-Studies/eeg_compression/ecomp_data/code/ecomp_analysis/" \ --overwrite=True done

The above will run the 01_find_bads.py script for each participant, using the overwrite option, and the data and analysis directory paths specified.

Note that some of the other scripts do not accept command line options. We recommend running them interactively with an editor such as VSCode (https://code.visualstudio.com/). VSCode will pick up on the # %% comments, and render parts of the script as code cells that can be individually executed.

General order to run code files

If you don't want to use the mpib_ecomp_derivatives dataset described above to skip some of the computations, you will need to execute all code files. The following list is a suggested order, because some code files will depend on others. Note that the data in the derived_data/annotations/ directory is crucial for reproducing results because parts of the information therein cannot be automatically provided (for example, which ICA components are rejected).

1. beh_misc.py
   1. two analyses will be skipped because they depend on outputs from later steps. Simply re-run this script after completing all other steps to get the respective results.
2. 00_sanity_check.py
3. 01_find_bads.py
4. 02_inspect_raw.py
5. 03_run_ica.py
6. 04_inspect_ica.py
7. 05_make_epochs.py
8. eeg_misc.py
9. erp_numbers.py
10. rdms_mahalanobis.py
11. rsa_timecourse.py
12. rsa_neurometrics.py
13. beh_accuracy.py
14. beh_modeling.py
    1. Run with different command line options: --fit_scenario=X, where X is (i) "free", (ii) "k_is_1", (iii) "k_smaller_1", and (iv) "k_bigger_1". Or set the fit_scenario variable in the script if you do not want to run this from the command line. (Keep fit_position set to "all")
    2. Run with different command line options fit_position=X, where X is a string "1", "2", ..., "10" (So you need to run this ten times). Additionally, run it with firsthalf and secondhalf. As above, you may also simply set the fit_position variable in the script. (Keep fit_scenario set to "free")
15. beh_weightings.py
    1. Run with different command line options: --fit_scenario=X, where X is (i) free, (ii) k_is_1. Or set the fit_scenario variable in the script if you do not want to run this from the command line.
16. plots_fig1.py
17. plots_fig2.py
18. plots_fig3.py
19. plots_fig4.py
20. Use LibreOffice Draw (https://www.libreoffice.org/discover/draw/) to open figures/paradigm_figure.odg and export paradigm_figure.pdf
21. Use xelatex (for example on a service like https://www.overleaf.com/) and figures/fig*.tex to finalize the figures. Note that the * stands for 1, 2, 3, or 4.

Explanation of files

General "infrastructure" files

• .flake8 --> code style configuration (see also pyproject.toml)
• .gitignore --> which files to ignore during version control
• .pre-commit-config.yaml --> for enforcing proper coding style
• .environment.yml --> software dependencies, see "Installation"
• pyproject.toml --> code style configuration (see also .flake8)
• README.md --> general information about this repository
• LICENSE --> how to use and credit the resources

Utility libraries

• config.py --> general configuration and variables that get re-used across scripts
• utils.py --> general functions that get re-used across scripts
• clusterperm.py --> functions for cluster-based permutation testing
• model_rdms.py --> script to generate model representational dissimilarity matrices

EEG preprocessing scripts

Generel preprocessing flow of the EEG data:

• 00_sanity_check.py
• 01_find_bads.py
• 02_inspect_raw.py
• 03_run_ica.py
• 04_inspect_ica.py
• 05_make_epochs.py

Miscellaneous analysis scripts

For example for checking for missing data or rejected EEG epochs:

• beh_misc.py
• eeg_misc.py

EEG analysis scripts

For both RSA and ERP analyses:

• rdms_mahalanobis.py
• rsa_neurometrics.py
• rsa_timecourse.py
• erp_numbers.py

Behavior analysis scripts

For general accuracy and descriptive behavior analyses, as well as modeling:

• beh_accuracy.py
• beh_modeling.py
• beh_weightings.py

Figure creation scripts

Each producing the figure as found in the manuscript (but see .tex files mentioned below):

• plots_fig1.py
• plots_fig2.py
• plots_fig3.py
• plots_fig4.py

Analysis outputs (derived_data/ directory)

This directory generally contains outputs from the scripts above. Importantly, the annotations subfolder contains important data from the visual screening of the EEG data (cannot be automatically reproduced):

• *_annotations.txt --> saved MNE-Python annotations
• *_bad-channels.txt --> each non-empty row in the file describes a bad channel
• *_bads_pyprep.json --> outputs from the 01_find_bads.py script
• *_exclude_ica.json --> which ICA components to exclude
• *_faster_bad_epos.json --> epoch indices rejected through the "FASTER" pipeline

Analysis outputs (figures/ directory)

This directory generally contains outputs from the scripts above. Exceptions are:

• paradigm_figure.odg --> to create fig1a, using LibreOffice Draw (https://www.libreoffice.org/discover/draw/)
• *.tex --> files that are used to put panel letters onto finished figures:
  • fig1.tex
  • fig2.tex
  • fig3.tex
  • fig4.tex