A simple wrapper script setup that facilitates running Kilosort2 on a batch cluster (e.g. PBS, SLURM). Also included in this repository are a number of scripts that wrap a number of common systems administration tasks (installing Kilosort2, testing the installation).
As per the kilosort2 install instructions you must have a version of MATLAB installed and a compatible version of CUDA. In our case, we used MATLAB 2018a with CUDA 9.0. Our particular configuration uses lmod to manage multiple versions of various applications. The installation script wrapper assumes the existence of two lmod modules: matlab/2018a and [email protected]. This installation script can be adapted for other setups, however.
First, clone the repository. Ensure that the git submodules are also cloned down:
cd /share/app
git clone [email protected]:ZuckermanBrain/kilosort2-cli.git
cd kilosort2-cli
git submodule update --init --recursive
To run mexGPUall.m with the appropriate version of CUDA and MATLAB, run the following script (replace /share/app with the path to where this repository is cloned):
/share/app/kilosort2-cli/bin/kilosort2-install
Then, install the environment module by moving it to a directory on your module path. If your MATLAB and CUDA modules are different from matlab/2018a and [email protected], edit the prereq statements in the TCL file before moving it.
mv /share/app/kilosort2-cli/modulefiles/kilosort2.tcl /share/modulefiles/manual/kilosort2
A script is included that will generate simulated eMouse drift data and then run it through the kilosort2 bash wrapper. To run it you must use a full or relative path (it will not be added to the path with the environment module):
/share/apps/kilosort2-cli/test/kilosort2-test
When it's done, it will print out paths to two test directories containing the results of the analysis and temporary files. These can be inspected if desired, or removed manually.
The usage is described in the dialogue that is presented when kilosort2 -h is run. Note that if lmod is used, pre-requisite modules must be loaded before kilosort2 is. Here is an example showing the modules being loaded, followed by an invocation of kilosort2 -h:
(base) [jsp2205@axon ~]$ ml load matlab/2018a
(base) [jsp2205@axon ~]$ ml load [email protected]
(base) [jsp2205@axon ~]$ ml load kilosort2
(base) [jsp2205@axon ~]$ kilosort2 -h
Usage: kilosort2 -[dtcmnseh]
=========================================================================
Description: Allows you to run Kilosort from GNU/Linux command line environment.
=========================================================================
All of the following command line options, followed by an argument, are required:
-d : The path to the directory that contains the raw data binary file.
-t : The path to temporary scratch space (same size as data, should be on a fast SSD).
-c : The path to your config file (.m format).
-m : The path to the channel map file (.mat format).
-n : The total number of channels in your recording.
-s : Start time of the time range to sort.
-e : End time of the time range to sort.
-h : Print this help message.
=========================================================================
Example usage: kilosort2 -d /data -t /tmp -c /home/jsp2205/configFile384.m -m /home/jsp2205/neuropixPhase3A_kilosortChanMap.mat -n 384 -s 0 -e Inf