Stability of motor representations after paralysis

This repository contains the scripts to generate figures for Stability of motor representations after paralysis.

Dataset

The dataset is available in NWB:N format from https://dandiarchive.org/dandiset/000147.

Analysis scripts

Python prerequisites

I recommend installing Python package dependencies using conda + poetry.

Install conda
Install poetry

Package dependencies

Create a minimal virtual environment:

conda create -n fingers_rsa python=3.8
conda activate fingers_rsa

We use poetry to specify package dependencies in pyproject.toml and poetry.lock. To install the dependencies, from the repository directory, run:

poetry install

Downloading the dataset

To download the dataset, change to the data directory and run the download-script:

cd data/
./download_data.sh

For non-Unix environments, you can download the dataset using the equivalent DANDI command:

dandi download DANDI:000147/0.220913.2243

Running the scripts

We use Hydra to specify configuration files and command-line arguments. For some scripts, the default configuration is already specified in the script, so you can run the script without any command-line arguments.

python scripts/peth.py
python scripts/crossval_classify.py

These scripts will output figures to an outputs/date/time/ directory (e.g. outputs/2022-10-13/11-18-05/). The scripts will also log their working/output directory to the console.

To calculate the representational dissimilarity matrices (RDMs), you need to specify the sessions to use for the RSA analysis.

python scripts/calc_rdm.py --multirun session=2018-09-10,2018-09-17a,2018-09-17b,2018-09-24,2018-09-26,2018-10-01,2018-10-12,2018-10-15,2018-10-17,2018-10-22

The following command is equivalent:

python scripts/calc_rdm.py -m +sweep=all_sessions

With the --multirun (-m) option, Hydra effectively runs calc_rdm.py 10 times, once for each session. Aligning and binning spikes takes some time, so you can parallelize multiruns using joblib with the additional command-line argument: hydra/launcher=joblib.

The calc_rdm.py multirun will generate outputs in multirun/date/time/run/ directories (e.g., outputs/2022-10-13/09-51-04/0). The output RDM absolute paths are also logged by calc_rdm.py, e.g.:

[2022-10-13 09:51:27,590][__main__][INFO] - Saving RDMs to: /home/user/Development/fingers_rsa/multirun/2022-10-13/09-51-04/session=2018-09-24/sub-P1_ses-2018-09-24_rdm.hdf5

The RDM files (absolute-path) are used as inputs to representational similarity analysis. For the example above, you would run:

python scripts/representational_similarity_analysis.py rdm_files="/home/user/Development/fingers_rsa/multirun/2022-10-13/09-51-04/*/*_rdm.hdf5"

For representational dynamics analysis (RDA), RDMs need to be calculated for multiple sessions and times. You can run:

python scripts/calc_rdm.py -m +sweep=all_sessions_and_times

This will take ~10 minutes even with parallelization.

Then, similar to RSA, you would pass in the resulting output RDM files to the RDA script:

python scripts/representational_dynamics_analysis.py rdm_files="/home/user/Development/fingers_rsa/multirun/date/time/*/*_rdm.hdf5"

Development environment

I developed and tested these Python scripts on Ubuntu 20.04 LTS. While they should work on other operating systems, I have not tested them. If you run into any issues, please open an issue and list your operating system and Python version.

Citation

Charles Guan, Tyson Aflalo, Carey Y Zhang, Elena Amoruso, Emily R Rosario, Nader Pouratian, Richard A Andersen (2022) Stability of motor representations after paralysis eLife 11:e74478. https://doi.org/10.7554/eLife.74478