Skip to content
This repository has been archived by the owner on Mar 15, 2023. It is now read-only.

Create NWB files by converting and combining neural data in proprietary formats and adding essential metadata.

License

Notifications You must be signed in to change notification settings

catalystneuro/nwb-conversion-tools

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PyPI version Full Tests Auto-release codecov documentation License

NWB Conversion Tools is deprecated. See NeuroConv instead.

NWB conversion tools (LEGACY)

NWB Conversion Tools is a package for creating NWB files by converting and combining neural data in proprietary formats and adding essential metadata.

Under heavy construction. API is changing rapidly.

Features:

  • Command line interface
  • Python API
  • Leverages SpikeExtractor to support conversion from a number or proprietary formats.

Installation

To install the latest stable release of nwb-conversion-tools though PyPI, type:

pip install nwb-conversion-tools

For more flexibility we recommend installing the latest version directly from GitHub. The following commands create an environment with all the required dependencies and the latest updates:

git clone https://github.com/catalystneuro/nwb-conversion-tools
cd nwb-conversion-tools
conda env create -f make_env.yml
conda activate nwb_conversion_env

Note that this will install the package in editable mode.

Finally, if you prefer to avoid conda altogether, the following commands provide a clean installation within the current environment:

pip install git+https://github.com/catalystneuro/nwb-conversion-tools.git@master

Dependencies

NWB Conversion Tools relies heavily on SpikeExtractors for electrophysiology and on ROIExtractors for optophysiology data.

You can use a graphical interface for your converter with NWB Web GUI.

Catalogue

v0.9.3

This project is an ongoing effort for the Ripple U19 conversion of extracellular electrophysiology data to NWB format, including final publishing of each dataset on DANDI. Currently spans 7 major publications and over 14 TB of data on the DANDI Archive. Most of the data consists of raw recordings, LFP, spike sorted units, and behavior with can consist of a mix of mental state tracking, position tracking through mazes, and trial stimulus events.

The Shenoy lab is one of the pioneers in developing BCIs for people with paralysis. They are part of the BrainGate team and were the winners of the 2019 BCI award. They use extracellular recordings from Utah arrays and Neuropixels in primates.

v0.9.2

The Brody lab has a long history with extracellular electrophysiology experiements spanning multiple acquisition systems. This project served two purposes - to allow the conversion of older data from Neuralynx and SpikeGadgets to NWB, and also their newer, larger data using Neuropixels (SpikeGLX). These recordings, some of which exceeded more than 250 GB (several hours worth!), were paired with rich trials tables containing catagorical events and temporal stimuli.

v0.8.10

The Feldman lab utilizes a Neuropixels (SpikeGLX) system along with multiple sophisticated behavior systems for manipulating whisker stimulation in mice. These give rise to very complex trials tables tracking multiple event times throughout the experiments, including multiple event trains within trials.

v0.8.1

Hussaini Lab: hussaini-lab-to-nwb

v0.7.2

v0.7.0

Neuropixel (SpikeGLX) recordings of subjects navigating a virtual reality! Behavior contains a huge variety of NWB data types including positional and view angle over time, collision detection, and more! Paired with a specific extension for parsing experiment metadata.

Utilizing the CED recording interface, this project paired ecephys channels with optogenetic stimulation via laser pulses, and mechnical pressure stimulation over time - all of which are channels of data extracted from the common .smrx files!

Other labs that use NWB standard

For Developers

Running GIN tests locally

nwb-conversion-tools verifies the integrity of all code changes by running a full test suite on short examples of real data from the formats we support. There are two classes of tests in this regard; tests/test_internals does not require any data to be present and represents the 'minimal' expected behavior for our package, whereas tests/test_on_data requires the user to both perform a full install of dependencies (pip install -r requirements-full.txt) as well as download the associated data for each modality.

Install testing dependencies

We provide two easy ways of installing all the dependencies required for testing:

  1. The first is a conda based solution that creates an environment with all the dependencies already installed.
git clone https://github.com/catalystneuro/nwb-conversion-tools
cd nwb-conversion-tools
conda env create -f make_env_testing.yml
conda activate nwb_conversion_testing_env

Note that this will also install datalad which is the endorsed way of downloading the testing data plus pytest and pytest-cov which are the tools that we use on our continuous integration suit.

  1. The same can be accomplished by using pip. In a clean environment run:
git clone https://github.com/catalystneuro/nwb-conversion-tools
cd nwb-conversion-tools
pip install .[test, full]

Notice that this method does not install datalad.

Downloading the data

Datalad (conda install datalad) is the recommended way for downloading the data. To do this; simply call:

For electrophysiology data:

datalad install -rg https://gin.g-node.org/NeuralEnsemble/ephy_testing_data

For optical physiology data:

datalad install -rg https://gin.g-node.org/CatalystNeuro/ophys_testing_data

For behavioral data:

datalad install -rg https://gin.g-node.org/CatalystNeuro/behavior_testing_data

Test configuration file

Once the data is downloaded to your system, you must manually modify the config file (example) located in ./tests/test_on_data/gin_test_config.json so its corresponding LOCAL_PATH key points to the correct folder on your system that contains the dataset folder (e.g., ephy_testing_data for testing ecephys). The code will automatically detect that the tests are being run locally, so all you need to do ensure the path is correct to your specific system.

The output of these tests is, by default, stored in a temporary directory that is then cleaned after the tests finish running. To examine these files for quality assessment purposes, set the flag SAVE_OUTPUTS=true in the gin_test_config.json file and modify the variable OUTPUT_PATH in the respective test if necessary.

Build the documentation

For building the documentation locally, the following procedure can be followed. Create a clean environment and type the following commands in your terminal:

git clone https://github.com/catalystneuro/nwb-conversion-tools
cd nwb-conversion-tools
pip install -e .[docs]

These commands install both the latest version of the repo and the dependencies necessary to build the documentation. Note that the argument -e makes you install editable

Now, to build the documention issue the following command in your terminal:

sphinx-build -b html docs ./docs/_build/

This builds the html under /docs/_build/ (from your root directory, where you have installed nwb-conversion-tools). This allows you to review the outcome of the process localy before commiting code.

About

Create NWB files by converting and combining neural data in proprietary formats and adding essential metadata.

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published