Spike Sorting Pipeline

This repository provides a reproducible pipeline for preprocessing, sorting, and analyzing spikes based on different feature sets (amplitude and widht, features from the SS-SPDF method Caro-Martín et al., 2018, raw waveform) via microneurography—using Snakemake and modular Python scripts. It also includes a Jupyter notebook for evaluating and visualizing classification results (e.g., heatmaps of accuracy).

The pipeline supports proprietary data formats like Dapsys and the process data format HDF5/NIX, converts them to a unified NIX format, and executes a full analysis including spike extraction (timestamps must be provided), feature set extraction, template computation, and classification evaluation.

Spike Tracking via the marking method

During the experiment the marking method is applied, a special electrical stimulation protocl to create spike tracks (vertical alignment of fiber responses). They can be extracted and analyzed post hoc.

In our workflow, we use two different tracking algorithms for microneurography data to identify and track spikes evoked by background stimuli:

• Dapsys (proprietary) – www.dapsys.net based on Turnquist et al., 2016
• SpikeSpy (open-source) – https://github.com/Microneurography/SpikeSpy

The extracted spike times and track labels are essential inputs for running our supervised spike sorting pipeline. The setup instructions are provided below.

---

Setup Instructions

1. Clone the repository

git clone https://github.com/Digital-C-Fiber/SpikeSortingPipeline.git
cd SpikeSortingPipeline

2. Create the Conda environment

This pipeline uses conda and snakemake with Python 3.11. We provide an environment file.

conda env create -f environment.yml
conda activate Snakemake311

If you haven't already installed snakemake:

conda install -c conda-forge snakemake

---

Snakemake Directory Overview

├── Snakefile                      # Main Snakemake workflow
├── config.yaml                    # Configuration for dataset paths and parameters
├── environment.yml                # Conda environment file
├── scripts/                       # Core processing scripts
│   ├── read_in_data.py
│   ├── preprocessing.py
│   ├── feature_extraction.py
│   ├── create_nix.py
│   ├── templates_and_filtering.py
│   ├── compute_template_errors.py
│   ├── classification.py
│   ├── clustering.py
│   └── collect_scores.py
│   └── test_clusters.py
├── datasets_test/
│   ├── testset_1.dps              # Example Dapsys file
│   └── nix/                       # Output folder for NIX

---

Configuration Format

Datasets and processing parameters are defined in config.yaml. Example:

datasets:
  Testset_1:
    path: "datasets_test/testset_1.dps"
    path_nix: "datasets_test/nix/testsets_1.nix"
    name: "testset_1"
    path_dapsys: "NI Puls Stimulator/Continuous Recording"
    flags:
      use_bristol_processing: false
    time1: 200
    time2: 922

Explanation:

• path: Raw data file (e.g. Dapsys)
• path_nix: Target output path for the .nix file
• path_dapsys: Root path inside the Dapsys hierarchy, usually it is NI Puls Stimulator/Continuous Recording or NI Pulse Stimulator/Continuous Recording, for h5/nix files just write ""
• use_bristol_processing: Adjusts preprocessing method for Bristol data
• time1, time2: Time window for analysis

---

Run the Pipeline

snakemake --cores 8

This will:

• Read raw data
• Create data frames
• Generate .nix files
• Apply preprocessing (align spikes, compute derivatives, and compute templates)
• Extract features
• Perform clustering/classification
• Output performance metrics for 6 implemented feature sets

---

Results Visualization

Use the provided notebook to plot and explore your results:

jupyter notebook visualize_results.ipynb

This notebook includes:

• Heatmaps of classification accuracy across feature sets
• Metric summaries (e.g. precision, recall, etc.)

---

Contact

If you have any questions, issues, or suggestions, feel free to reach out:

Alina Troglio Email: [email protected]

---

How to Cite

If you use this pipeline in your work, please cite our preprint:

Supervised Spike Sorting Feasibility of Noisy Single-Electrode Extracellular Recordings: Systematic Study of Human C-Nociceptors recorded via Microneurography