Model-based spike sorting with a mixture of drifting t-distributions

Area of Science:

• Systems Neuroscience
• Computational Neuroscience

Background:

• Chronic extracellular recordings are vital for systems neuroscience.
• Spike sorting remains a significant challenge in data analysis.
• Generative models, like Gaussian mixtures, quantify unit isolation quality.

Purpose of the Study:

• To introduce a novel spike sorting strategy for chronic extracellular recordings.
• To address limitations of existing methods, particularly cluster drift and outliers.
• To provide a robust measure of unit isolation quality.

Main Methods:

• Modeling spike data using a mixture of drifting t-distributions.
• Capturing temporal cluster drift and heavy-tailed spike distributions.
• Developing a software implementation for rapid, interactive clustering.

Main Results:

• The drifting t-distribution model fits empirical data better than Gaussian mixtures.
• The model demonstrates improved robustness to outliers.
• Accurate estimation of misclassification error compared to existing metrics.

Conclusions:

• The mixture of drifting t-distributions model facilitates efficient spike sorting of extensive datasets.
• It offers a reliable measure of unit isolation quality across diverse recording conditions.
• Enables interactive analysis of chronic neural recordings.