Efficient sorting of single-unit activity from midbrain cells using KiloSort is as accurate as manual sorting.

Extracting single-unit activity from in vivo extracellular neural electrophysiology data requires sorting spikes from background noise and action potentials from multiple cells in order to identify the activity of individual neurons. Typically this has been achieved by algorithms that employ principal component analyses followed by manual allocation of spikes to individual clusters based on visual inspection of the waveform shape. This method of manual sorting can give varying results between human operators and is highly time-consuming, especially in recordings with higher levels of background noise. To address these problems, automatic sorting algorithms have begun to gain popularity as viable methods for sorting electrophysiological data, although little is known about the use of these algorithms with neural data from midbrain recordings. KiloSort is a relatively new algorithm that automatically clusters raw data which can then be manually curated. In this report, we compare results of manually-sorted and KiloSort-processed recordings from the ventral tegmental area (VTA) and substantia nigra pars compacta (SNc). Sorting with KiloSort required substantially less time to complete, while yielding comparable and consistent results. We conclude that the use of KiloSort to identify single units from multi-channel recording in the VTA and SNc is accurate and efficient.