A scaling cross platform tool for the analysis of neurophysiological data

Abstract

This paper describes the development of a cross platform crosscorrelation and clustering application for neural spike train recordings that will scale seamlessly from use on a researchers PC to a high performance computing cluster (HPC). Clusters of neurons are identified from the neural recordings using a hierarchical agglomerative clustering algorithm applied to the cross correlation data. Finally a cross correlation grid is used to visualise the neural clusters with the user navigating through the dataset using a dendrogram depicting the identified clusters. The cross correlation algorithm is an “embarrassingly parallel problem” that is scaled to cope with a large number of neurons through exploiting multiple compute cores both in the setting of a researchers PC or an HPC. This scaling is achieved using MPJ Expressa Java implementation of the Message Passing Interface (MPI).