phy_extras

A collection of phy enhancements building on Allen Institute's spikesorting metrics among others

Features

1. Trace view filtering and referencing

2. Auto-split via K-means, GAC, or time

Installation

1. Install development versions of phy and phylib if you haven't yet:

# Conda install dependencies 
conda install -y numpy matplotlib scipy h5py dask cython pillow colorcet pyopengl requests qtconsole tqdm joblib click toolz scikit-learn
pip install git+https://github.com/mmyros/ssm.git

2. do one of the two things:

• Copy dotphy directory to your home directory and rename it .phy
• OR, on Mac or Linux, make a symbolic link: ln -s ~/phy_extras/dotphy/ ~/.phy (assuming you downloaded this repository to your home directory)

Usage

Next time phy is launched, new columns should show up in cluster metrics. Also, the following keyboard shortcuts should work:

1. Alt+R to cycle through raw data and waveform views: raw, highpass, mean+highpass, median+highpass, highpass+median, highpass+mean
2. s for Scrub outliers using Local Outlier Factor (also implemented are Robust covariance, One-Class SVM, and Isolation Forest. See plugins/custom_split.py)
3. t for Time split using gaussian hierarchical model
4. x for k-means split
5. d for GAC split (Due to Swindale lab, gradient ascent clustering (GAC) algorithm, a variant of the mean-shift algorithm)

Finally, the file cluster_metrics.csv will appear and contain the calculated cluster quality metrics

plugins/spike_io.py can be run standalone from the terminal or Anaconda prompt. It will generate cluster_metrics.csv independently of phy

Clustering algorithms tested

Tested on Debian Linux with:

• Matlab-based Kilosort 1 and
• python-based Pykilosort 2.

Should also work with Matlab-based Kilosort 2 but untested

Warning

Splitting routines are meant to be used as exploratory tools to enhance understanding of clustering. I hope you find them useful to refine your core clustering algorithm's parameters