XSort

A Python/Qt application that visualizes and edits the results from spike sorting of neuronal activity recorded with extracellular multi-electrode probes during experiments conducted in the Stephen G. Lisberger lab at Duke University.

Background

Researchers in the Lisberger and other neuroscience lab use multi-electrode recording systems like the Omniplex Neural Recording Data Acquisition System to record electrical activity in the brain during experiments. They analyze the recorded analog channel data streams to detect and "sort" the electrical activity into distinct "spike trains", each assigned to a different "neural unit".

The various spike-sorter algorithms employed in the lab often produce spurious results -- "garbage" units triggered by system noise, two spike trains that have a very similar spike waveform and really "belong" to the same neural unit, and so on. With XSort you can visually assess and edit the original spike-sorter output.

To use the application, select a working directory containing the analog data recording (an Omniplex.pl2 file, or a supported flat binary file format) and the original spike sorter results (a Python pickle file, .pkl). After analying the files and building an internal cache of channel data streams and unit metrics (in individual files stored in the working directory), XSort offers a number of different data visualizations.

1. A tabular listing of the neural units, including metrics -- total # of spikes observed, mean firing rate, best signal-to-noise ratio (across all data channels), the primary data channel (on which best SNR was measured), and others.
2. Channel traces for each of up to 16 analog data channels, highlighting the spikes of each of up to 3 units currently selected for display in the neural unit list, known as the display list.
3. The per-data channel spike template waveforms (10ms, 1ms pre-spike time) for each unit in the display list.
4. A firing rate-vs-time histogram, inter-spike interval histogram, and autocorrelogram for each unit in the display list, along with the cross-correllogram of each unit in the display list vs another unit in the list.
5. A 3D histogram (rendered as a heatmap) representing the unit's autocorrelogram as a function of instantaneous firing rate.
6. A principal component analysis (PCA) scatter plot projecting s a unit's spikes into a 2D space defined by the two highest-variance (most information) principal components. When the analysis is performed on multiple units, it can help to detect distinct units that are really part of the same unit. When performed on a single neural unit, it can help detect a unit that should be split into two or more distinct spike populations.

You can edit the list of neural units in a number of ways:

1. Attach a short descriptive label (typically, the putative neuron type) to any unit.
2. Delete a selected unit.
3. Merge two selected units into one.
4. Split the currently selected unit into two distinct units.

XSort maintains a complete "edit history" in the current working directory, and you can "undo" any or all of the changes in reverse order, or undo them all at once to recover the original state of the neural list. Once happy with any changes made, you can save the neural list for use in other applications or your own scripts.

Installation (for MacOS/Linux)

• Ensure that Python 3.9+ is installed on your system. We currently build the package against version 3.11.4. Contact the developer if you run into any issues running the program on 3.9.x or 3.10.x.
• Download the wheel file xsort-x.y.z-py3-none-any.whl, where x.y.x is the release version number.
• In a terminal console, navigate to the directory holding the wheel file you downloaded, and install the package: pip install xsort-x.y.z-py3-none-any.whl.
• To start the app: python -m xsort.app.
• A command-line no-GUI version of XSort is now available that builds the internal cache for a specified working directory. For detailed usage information: python -m xsort.app ?.
• You may run into the following error when starting XSort: qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "" even though it was found. This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem. Available platform plugins are: minimalegl, minimal, wayland-egl, vkkhrdisplay, offscreen, eglfs, vnc, linuxfb, xcb, wayland. If so, try reinstalling XCB platform plugin with sudo apt-get install '*libxcb*'
• The XSort wheel file is configured to require Python 3.9 or later. You may want to avoid the "latest and greatest" Python release as issues may arise. Eg, pip could not install Numpy <= 1.25 when the installed Python version was 3.12. The Numpy developers released 1.26 to address the problem. A change in the wheel configuration as of XSort 0.1.4 should address the conflict and ensure that the right version of Numpy is installed if the Python version is 3.12 or later.
• XSort takes advantage of Python's multiprocessing library to speed up the background work it does to build a set of internal cache files (for optimal data retrieval) and to calculate neural unit statistics displayed in its different views. It is highly recommended that you run the program on a multi-core system -- the more cores the better!

License

XSort was created by Scott Ruffner. It is licensed under the terms of the MIT license.

Credits

Based on a program developed by David J Herzfeld, who provided critical guidance and feedback during the development of XSort. Developed with funding provided by the Stephen G. Lisberger laboratory in the Department of Neurobiology at Duke University.

In addition to the Python standard library, the XSort user interface is built upon the Qt for Python framework, PySide6, data analysis routines employ the Numpy and SciPy libraries, and graphical plots are drawn using PyQtGraph.