Automated spike sorting curation

Identifying spike timing of individual neurons recorded simultaneously is essential for investigating the neuronal circuits that underlie cognitive functions. Recent developments allow the simultaneous recording of multi-channel extracellular spikes from hundreds of neurons. Yet classifying extracellular spikes into putative source neurons remains a challenge. Here, we set out to develop a fully automated algorithm which can replace manual curation for an existing multi-channel spike sorting solution. As an input to the automated curator (AC), we use KlustaKwik to over-cluster the spikes automatically, yielding numerous spike clusters. From that point onwards, we refer to spike sorting not as a clustering problem, but rather as a sequence of two binary classification problems. To perform automatic curation, we first use a “noise classifier” to remove non-neuronal clusters. Second, we use a “merge classifier” to determine whether any given pair of clusters should be merged. The output of the AC is a set of well-isolated units