This repository contains the code of automated spike sorting curation, recently presented as “Multi-channel automated spike sorting achieves near-human performance”, in FENS forum 2022, Paris, France, July 2022.

Identifying spike timing of individual neurons recorded simultaneously is essential for investigating the neuronal circuits that underlie cognitive functions. Recent developments allow the simultaneous recording of multi-channel extracellular spikes from hundreds of neurons. Yet classifying extracellular spikes into putative source neurons remains a challenge. Here, we set out to develop a fully automated algorithm which can replace manual curation for an existing multi-channel spike sorting solution. As an input to the automated curator (AC), we use KlustaKwik to over-cluster the spikes automatically, yielding numerous spike clusters. From that point onwards, we refer to spike sorting not as a clustering problem, but rather as a sequence of two binary classification problems. To perform automatic curation, we first use a “noise classifier” to remove non-neuronal clusters. Second, we use a “merge classifier” to determine whether any given pair of clusters should be merged. The output of the AC is a set of well-isolated units.

Dataset: The input for the automated curator should be a dataset in a neurosuite format (Hazan et al., 2006). Specifically, the following three files are required:

• clu: spike labels, e.g. session_name.clu.1
• res: spike times in samples, e.g. session_name.res.1
• spk: spike waveforms, e.g. session_name.spk.1

The current version of the AC works only with data which were recorded at 20 kHz, and spike waveforms which are 32 samples long.

The code was tested on machines running Linux (Ubuntu 18), but is expected to work on any OS

• Conda (both Miniconda and Anaconda are fine)
• Python>3.6

For installing the automated curator, first create an activate a virtual environment using conda:

$ conda create --name myenv
$ conda activate myenv

Second, install the FastAi package, if you are using Miniconda run:

conda install -c fastchan fastai

or if you are using Anaconda then run:

conda install -c fastchan fastai anaconda

Third, install the Phylib package with pip:

$  pip install phylib

Finally, clone the repository to your machine:

$ git clone https://github.com/EranStarkLab/Automated-curation.git

• Open the running_AC_pipeline.py file in your Python IDE
• Change lines 24-26 according to your dataset which you want to curate
• Run the running_AC_pipeline.py script

• Navigate to the automated curator folder

$ cd filebase/to/automated/curator/folder

• Activate conda virtual environment

$ conda activate myenv

• Run the automated curator script with three inputs:

1. Filebase: full path to session dataset with session name, e.g. ~/home/data/session_name
2. shank_number: shank number, e.g. 1
3. n_channels: number of channels in the spk file, e.g 10

$ python running_AI_pipeline_bash.py Filebase n_shanks n_channels

The output of the automated curator is a list of lists. Every list is a recommendation for an action, that can be carried out in the environment of klusters. The first list (index 0) contains the “Noise/Multi-units”, clusters which are either non-neuronal units or multi-neuronal units which cannot be separated into single units. Every other list (index 1 and higher) contains two or more units which should be merged with one another.