AutoCurationKilosort: Automated Curation for Kilosort Output

A tool designed to assist researchers in automating the curation process of Kilosort outputs. It helps streamline the curation of sorting results, reducing manual effort while maintaining control over the final output for increased accuracy and trustworthiness.

How to use it

• Add the path to the repository to the MATLAB path

addpath(genpath('path_to_AutoCurationKilosort'))

• Edit the settings.json file to set the curation parameters

• Copy AutoCurationKilosort.m to the your data folder. Edit the installation path in the script.

folder_data = './catgt_Exp_g0';
setting_filenames = 'path_to_AutoCurationKilosort/settings.json';

• Run the script AutoCurationKilosort.m

The pipeline

• Remove the clearly bad units with defined SNR thresholds
• Remove the outlies inside each cluster in the PC-feature space
• Find the potential splits and merges and make the dicisions automatically / manually
• Compute the quality metrics for each unit
• Label the units as 'good' or 'mua' based on the quality metrics
• Center the troughs of the waveforms to the spike times

Why should we use it

• Do not waste time on the clearly bad units (noise)
• It is tiring to look at thousands of units, after you have gained the clear understanding of the data
• It is laborious to curate the waveforms of each unit by remove the large noise in the waveforms
• It is error-prone and inconsistant to label the units as 'good' or 'mua' manually

Documentation

See Documentation.md for details.

References for code

Quility metrics
Sean Bone (2024). JSON+C parsing for MATLAB (https://github.com/seanbone/matlab-json-c/releases/tag/v1.1), GitHub.
Algorithm AS 217 APPL. STATIST. (1985) Vol. 34. No.3 pg 322-325