📝 SummarXiv

Abstract

Clustering algorithms became an essential part of the neurophysiological data analysis toolbox in the last twenty five years. Many problems, from the definition of cell types/groups based on morphological, molecular and physiological data to the identification of sub-networks in fMRI data, are now routinely tackled with clustering analysis. Since the datasets to which this type of analysis is applied tend to be defined in larger and larger dimensional spaces, there is a need for efficient and robust clustering methods in high dimension. There is also a need for methods that assume as little as possible about the clusters shape and size. We report here our experience with the ToMATo (Topological Mode Analysis Tool) algorithm. It is based on a definitely deep mathematical theory (algebraic topology), but its Python based open-source implementation is easily accessible to practitioners. We applied ToMATo to a problem we know well, spike sorting. Its capability to work in the ''native'' space of the data (no dimension reduction is required) is remarkable, as well as its robustness with respect to outliers (superposed spikes).