Spike sorting of heterogeneous neuron types by multimodality-weighted PCA and explicit robust variational Bayes

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Accurate spike sorting is crucial for understanding neural population dynamics.
• Existing methods struggle with heterogeneous neural populations and difficult spike patterns (e.g., bursting or sparsely firing neurons).
• There is a need for robust and efficient spike sorting algorithms.

Purpose of the Study:

• To introduce a novel spike sorting method for classifying spike waveforms from multiunit recordings.
• To develop a method capable of sorting spike mixtures from heterogeneous neural populations.
• To improve the accuracy and reliability of spike sorting, especially for challenging neural data.

Main Methods:

• Feature extraction using multimodality-weighted principal component analysis (mPCA).
• Clustering using an explicit variational Bayes for Student's t mixture model (SVB) without Maximum-A-Posterior (MAP) inference.
• Comparison of the proposed method against conventional techniques using simulated and experimental datasets.

Main Results:

• mPCA effectively extracts informative features, creating separable clusters in a low-dimensional space.
• The explicit SVB implementation is faster and more reliable than conventional SVB, particularly for difficult-to-sort spike patterns.
• The proposed method demonstrated significantly improved performance in sorting spikes from bursting and sparsely firing neurons.

Conclusions:

• The combined mPCA and explicit SVB approach offers a significant advancement in spike sorting accuracy and efficiency.
• This method reliably handles complex neural data, including signals from heterogeneous populations and difficult neuron types.
• The open-source implementation (EToS version 3) facilitates broader adoption and further research in neural signal analysis.