Efficient and reproducible pipelines for spike sorting large-scale electrophysiology data

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Electrophysiology

Background:

• Large-scale in vivo electrophysiology is expanding, enabling new discoveries but demanding significant computational resources.
• Spike sorting, a crucial step for analyzing electrophysiology data, is a computational bottleneck, hindering the full potential of high-density recordings.
• Current methods lack standardized validation, making it difficult to compare different spike sorting algorithms and preprocessing techniques.

Purpose of the Study:

• To develop a scalable, end-to-end spike sorting pipeline that leverages parallelization for efficient processing of large electrophysiology datasets.
• To introduce a parallelized benchmarking pipeline for systematic and reproducible comparison of spike sorting algorithms and data compression strategies.
• To address the need for faster, more accurate, and transparent spike sorting solutions for the growing field of large-scale electrophysiology.

Main Methods:

• Developed an end-to-end spike sorting pipeline utilizing parallel processing for scalability across workstations, HPC clusters, and cloud environments.
• Implemented a parallelized benchmarking pipeline to systematically compare multiple spike sorting algorithms and preprocessing steps.
• Evaluated the performance of Kilosort4 against Kilosort2.5 and assessed the impact of 7x lossy compression on spike sorting accuracy.

Main Results:

• The developed spike sorting pipeline successfully scales to large electrophysiology datasets, reducing processing time and costs.
• Kilosort4 demonstrated superior performance compared to Kilosort2.5 within the established benchmarking framework.
• 7x lossy compression was found to have a minimal impact on spike sorting performance, offering significant data storage savings.

Conclusions:

• The presented parallelized spike sorting and benchmarking pipelines provide essential tools for handling the computational demands of modern electrophysiology.
• These scalable and transparent solutions facilitate rigorous validation and comparison of spike sorting methods.
• The findings support the adoption of efficient data processing and storage strategies for future multi-thousand-channel electrophysiology experiments.