A multistage mathematical approach to automated clustering of high-dimensional noisy data

Source abstract

Significance Organizing large, multidimensional datasets by subgrouping data as clusters is a major challenge in many fields, including neuroscience, in which the spike activity of large numbers of neurons is recorded simultaneously. We present a mathematical approach for clustering such multidimensional datasets in a relatively high-dimensional space using as a prototype datasets characterized by high background spike activity. Our method incorporates features allowing reliable clustering in the presence of such strong background activity and, to deal with large size of datasets, incorporates automated implementation of clustering. Our approach effectively identifies individual neurons in spike data recorded with multiple tetrodes, and opens the way to use this method in other domains in which clustering of complex datasets is needed.