Application of Parallel Factor Analysis (PARAFAC) to electrophysiological data.

S Katharina Schmitz,Ralf A W Galuske,Boris Ebisch,Anja Klein,Gordon Pipa,Philipp P Hasselbach

doi:10.3389/fninf.2014.00084

Abstract

The identification of important features in multi-electrode recordings requires the decomposition of data in order to disclose relevant features and to offer a clear graphical representation. This can be a demanding task. Parallel Factor Analysis (PARAFAC; Hitchcock, 1927; Carrol and Chang, 1970; Harshman, 1970) is a method to decompose multi-dimensional arrays in order to focus on the features of interest, and provides a distinct illustration of the results. We applied PARAFAC to analyse spatio-temporal patterns in the functional connectivity between neurons, as revealed in their spike trains recorded in cat primary visual cortex (area 18). During these recordings we reversibly deactivated feedback connections from higher visual areas in the pMS (posterior middle suprasylvian) cortex in order to study the impact of these top-down signals. Cross correlation was computed for every possible pair of the 16 electrodes in the electrode array. PARAFAC was then used to reveal the effects of time, stimulus, and deactivation condition on the correlation patterns. Our results show that PARAFAC is able to reliably extract changes in correlation strength for different experimental conditions and display the relevant features. Thus, PARAFAC proves to be well-suited for the use in the context of electrophysiological (action potential) recordings.

Highlights

Action potentials are the means of information transmission between neurons
To demonstrate the application of Parallel Factor Analysis (PARAFAC) to multiunit spiking data, we applied the algorithm to an example dataset obtained in the experiment described above
The first 21 repetitions of each stimulus were recorded without deactivation, repetitions 22–42 correspond to the phases of thermal deactivation of posterior middle suprasylvian (pMS) and repetitions 43–63 show the results for the rewarm condition

Summary

Introduction

Action potentials are the means of information transmission between neurons. Synchronous firing of action potentials is believed to be one of the crucial mechanisms of information coding in the brain (for reviews see Singer, 1999; Uhlhaas et al, 2009). The examination of the temporal structure of spike trains (sequences of action potentials) and the detection of correlations among the signals of multi-electrode recordings can provide fundamental insights into presumptive coding strategies. To this end, the temporal structure in parallel recordings of spike trains has been used to assess the flow of information among neurons (Grün et al, 2002; Pipa et al, 2008)

Methods

Results

Conclusion