YASS

YASS performs spike sorting from large-scale multi-electrode array (MEA) recordings to extract and deconvolve neuronal spike events and improve accuracy and robustness in challenging datasets such as primate retina.

Key Features:

• Efficient triage system: Neural network-based detection and denoising identify clean, high-quality waveforms while temporarily excluding noisy or collided events.
• Outlier triaging: An outlier triage step removes spurious signals to improve the dataset used for clustering.
• Nonparametric Bayesian clustering: Nonparametric Bayesian clustering infers spike templates from denoised waveforms without presetting the number of clusters.
• Parallelized divide-and-conquer strategy: A divide-and-conquer approach parallelizes clustering to scale to large datasets.
• Matching-pursuit deconvolution: Matching-pursuit deconvolution recovers collided waveforms to disentangle overlapping spikes.
• Split-and-merge template refinement: Split-and-merge steps refine and estimate additional templates from recovered waveforms.

Scientific Applications:

• Primate retina MEA recordings: Applied to primate retina data to handle high firing rates and overlapping axonal units.
• Collision recovery: Enables recovery and separation of collided spikes arising from synchronous neuronal firing.
• Quality assessment via receptive-field mosaics: Uses the well-defined mosaic structure of receptive fields as a quality check for spike sorting output.
• High-density MEA benchmarks: Demonstrated superior performance on real and semi-simulated MEA datasets with more than 500 electrodes compared to state-of-the-art and manual sorting methods.

Methodology:

Implements a triage-then-cluster-then-pursuit pipeline integrating neural network-based detection and denoising, outlier triaging, nonparametric Bayesian clustering, divide-and-conquer parallelization, matching-pursuit deconvolution, and split-and-merge template refinement.