AUTOMATIZED CURATION OF SPIKE SORTING CLUSTERS
1 Forschungszentrum Jülich
The availability of high-density electrophysiology techniques is transforming systems neuroscience, enabling experimental labs to record activity from hundreds of neurons throughout the brain simultaneously. However, extracting single-cell activity from the resulting large datasets remains challenging, as various experimental conditions introduce sources of contaminating noise, and even optimized spike-sorting algorithms may produce false positive units. This requires labor-intensive manual evaluation of spike clusters.To address this challenge, we extended existing analyses and incorporated machine learning to reduce the need for manual data curation. Using Neuropixels probes, we collected diverse datasets, including recordings of varying quality from multiple experimental setups and recordings with optogenetic stimulation or functional imaging. We hand-labeled resulting spiking clusters to generate supervised machine-learning labels. Our model features consist of both existing and new quality metrics, such as synchrofacts, firing rate drift, and amplitude variation, and the approach utilizes a classifier to identify noise clusters automatically. Furthermore, we successfully extended our approach to separate well-isolated single-cell clusters from mixed-population activity. Our package is compatible with the Spikeinterface API, facilitating broader adoption. We found that our approach generalizes across recording sessions, animals, and brain areas, and applies to low-density probes used in human epilepsy patients.
Our study highlights the importance of evaluating high-density spike sorting outputs and presents an automated processing pipeline, including novel methods for data denoising and classification. This approach is user-friendly and extendable with additional metrics, thus providing a powerful tool to efficiently isolate single-cell activity from large datasets.