Klusta: automatic spike sorting up to 64 channels

klusta is an open source package for automatic spike sorting of multielectrode neurophysiological recordings made with probes containing up to a few dozens of sites.

We are also working actively on more sophisticated algorithms that will scale to hundreds/thousands of channels. This work is being done within the phy project, which is still experimental at this point.

Overview

klusta implements the following features:

• Kwik: An HDF5-based file format that stores the results of a spike sorting session.
• Spike detection (also known as SpikeDetekt): an algorithm designed for probes containing tens of channels, based on a flood-fill algorithm in the adjacency graph formed by the recording sites in the probe.
• Automatic clustering (also known as Masked KlustaKwik): an automatic clustering algorithm designed for high-dimensional structured datasets.

GUI

You will need a GUI to visualize the spike sorting results.

We have developed two GUI programs with the same features:

• phy KwikGUI: newer project, scales to hundreds/thousands of channels, still relatively experimental. It will be automatically installed if you follow the install instructions below.
• KlustaViewa: widely used, but older and a bit hard to install since it relies on very old dependencies.

Both GUIs work with the same Kwik format.

Quick install guide

The following instructions will install both klusta and the phy KwikGUI.

1. Make sure that you have miniconda installed. You can choose the Python 3.5 64-bit version for your operating system (Linux, Windows, or OS X).

2. Download the environment file.

3. Open a terminal (on Windows, cmd, not Powershell) in the directory where you saved the file and type:

   conda env create -n klusta -f environment.yml

4. Done! Now, to use klusta and the phy KwikGUI, enter the directory that contains your files and type:

   source activate klusta  # omit the `source` on Windows
   klusta yourfile.prm  # spikesort your data with a PRM file
   phy kwik-gui yourfile.kwik  # open the GUI

   See the documentation for more details.

Updating the software

To get the latest version of the software, open a terminal and type:

source activate klusta  # omit the `source` on Windows
pip install klusta phy phycontrib --upgrade

Technical details

klusta is written in pure Python. The clustering code, written in Python and Cython, currently lives in another repository.

Links

• Documentation (work in progress)
• Paper in Nature Neuroscience (April 2016)
• Mailing list
• Sample data repository (work in progress)

Credits

klusta is developed by Cyrille Rossant, Shabnam Kadir, Dan Goodman, Max Hunter, and Kenneth Harris, in the Cortexlab, University College London.