RT-Sort

RT-Sort: an action potential propagation-based algorithm for real time spike detection and sorting with millisecond latencies

Source code organization

This repository contains all of the code used in the development of RT-Sort and used to create the figures in the RT-Sort publication. The source code is divided into 3 main folders. The folders and some highlighted scripts are described below:

1. spike_detection: Training and validating the DL models used for spike detection.
   • spikesort_matlab4.py: Spike sort (with Kilosort2) and curate recordings used for training and validating the DL model
   • train/single_mea.ipynb and train/single_neuropixels.ipynb: The core training and validating of the MEA and Neuropixels models, respectively.
2. spike_sorting: The sequence detection and spike assignment portion of RT-Sort (and some code for validating the DL models).
   • run_alg/si_rec13: Jupyter Notebooks that start with this run the complete RT-Sort algorithm (from sequence detecting to simulated real time spike sorting) on a specific recording or group of recordings.
                      si_rec13_patch.ipynb: Ground truth MEA
                      si_rec13_ground_truth.ipynb: Simulated mouse in vivo Neuropixels
                      si_rec13.ipynb: Mouse in vivo Neuropixels
3. additional_analysis: Additional analyses not included in spike_detection or spike_sorting, such as Figure 4B-G.

Installation

In terminal with Anaconda installed, run conda env create -f env.yml to create the Python environment with all the libraries with the correct versions installed.

If that does not work, install the following Python libraries (ideally with the specified versions):

• h5py==3.7.0
• numpy==1.21.6
• matplotlib==3.5.3
• scipy==1.7.3
• scikit-learn==1.0.2
• torch==1.12.1+cu113
• torch_tensorrt==1.2.0
• diptest==0.6.0
• tqdm==4.65.0
• spikeinterface==0.94.0
• jupyter==1.0.0
• comet-ml==3.36.0

Running source code

Many scripts act as modules for other scripts. For the scripts in src/spike_detection, src used in import statements refers to the folder src/spike_detection. For the scripts in src/spike_sorting, src used in import statements refers to the folder src/spike_sorting.

Terminology discrepancies

There are some discrepancies in the terminology used in the source code and the RT-Sort publication. The following are the most notable:

• Variations of propagation signal, such as prop_signal, refer to the RT-Sort algorithm.
• latencies and delays in the source code refer to "time intervals" and "latencies" in the RT-Sort publication, respectively.
• clusters may refer to "preliminary propagation sequences" depending on the context.
• SI or si refer to Neuropixels.

Correspondence

Tjitse van der Molen: [email protected]
Max Lim: [email protected]