2020 Theses Doctoral

Spike Sorting for Large-scale Multi-electrode Array Recordings in Primate Retina

Lee, Jin Hyung

Spike sorting is a critical first step in extracting neural signals from large-scale multi-electrode array (MEA) data. This manuscript presents several new techniques that make MEA spike sorting more robust and scalable.

The first part explains the methods and the evaluations of YASS (Yet Another Spike Sorter), a pipeline designed for MEA spike sorting. Our pipeline is based on an efficient multi-stage ``triage-then-cluster-then-pursuit'' approach that initially extracts only clean, high-quality waveforms from the electrophysiological time series by temporarily skipping noisy or ``collided'' events (representing two neurons firing synchronously).

This is accomplished by developing a neural network detection and denoising method followed by efficient outlier triaging. The denoised spike waveforms are then used to infer the set of spike templates through nonparametric Bayesian clustering. We use a divide-and-conquer strategy to parallelize this clustering step.

Finally, we recover collided waveforms with matching-pursuit deconvolution techniques, and perform further split-and-merge steps to estimate additional templates from the pool of recovered waveforms. We apply the new pipeline to data recorded in the primate retina, where high firing rates and highly-overlapping axonal units provide a challenging testbed for the deconvolution approach; in addition, the well-defined mosaic structure of receptive fields in this preparation provides a useful quality check on any spike sorting pipeline.

We find that the proposed methods improve on the state-of-the-art on both real and semi-simulated MEA data with >500 electrodes. The second part discusses a novel approach to spike sorting using the Neural Clustering Process (NCP), a recently introduced neural architecture that performs scalable amortized approximate Bayesian inference for efficient probabilistic clustering. To optimally encode spike waveforms for clustering, we extended NCP by adding a convolutional spike encoder, which is learned end-to-end with the NCP network.

Trained purely on labeled synthetic spikes from a simple generative model, the NCP spike sorting model shows promising performance for clustering multi-channel spike waveforms. The model provides higher clustering quality than an alternative Bayesian algorithm, finds more spike templates with clear receptive fields on real data, and recovers more ground truth neurons on hybrid test data compared to a recent spike sorting algorithm. Furthermore, NCP is able to handle the clustering uncertainty of ambiguous small spikes by GPU-parallelized posterior sampling.