Neuronal Spike Waveform Clustering

MATLAB and Python workflows for exploring neuronal spike waveforms, calculating waveform attributes, comparing k-means with Ward hierarchical clustering, and visualizing clusters in PCA space.

Example results

The repository also includes a plot of all waveforms and a results slide deck.

Methods

• Inspect waveform traces and check for missing values
• Calculate maximum amplitude, minimum amplitude, and area under the curve
• Classify waveforms by the relative size of their positive and negative peaks
• Group waveforms using k-means and Ward hierarchical clustering
• Project waveforms onto the first two principal components to visualize cluster assignments
• Plot individual waveforms and mean traces for each cluster

PCA is used for visualization in these scripts; clustering is performed on waveform samples. Area under the curve is calculated as a waveform attribute, not used as a clustering input.

Repository contents

• analyze_spike_waveforms.m — MATLAB waveform attributes, clustering, PCA plots, cluster traces, and attribute plots
• spike_clustering.py — Python command-line workflow that saves waveform, PCA, and cluster-trace figures
• neuroPixels_crossCorr_night.m — separate MATLAB cross-correlation analysis
• DS.txt — optional direction-selectivity values; the Python clustering script checks this file but does not use it to form clusters
• *.png — selected example outputs
• spike_clustering_results_verified.pptx — additional results

Input data

The clustering scripts require a text file containing one waveform per row and one sample per column. spike_waveforms.txt is not currently included in this repository. Supply an appropriately shareable waveform file before running the scripts.

The MATLAB script expects spike_waveforms.txt beside analyze_spike_waveforms.m. It reads DS.txt from the same location when available. It stops if the waveform matrix contains NaN values, so clean missing values before running it.

Run the MATLAB analysis

Place spike_waveforms.txt in the repository directory and run analyze_spike_waveforms.m in MATLAB. The script uses eight clusters by default; change numClusters near the top to use another value.

Run the Python analysis

Install NumPy, Matplotlib, SciPy, and scikit-learn, then run:

python spike_clustering.py spike_waveforms.txt --output-dir figures

To include the optional direction-selectivity file:

python spike_clustering.py spike_waveforms.txt --ds DS.txt --output-dir figures

The Python script uses the first 2,000 waveforms for its clustering demonstration by default. Adjust this with --n-demo; adjust the cluster count with --clusters. Ward clustering can be skipped with --no-ward.

Related work

Data_Science_Tasks contains Jupyter notebooks for PCA, clustering-method comparison, and cluster characterization. neuropixels_analysis contains a separate waveform import and cleaning notebook.