Neuronal Assembly Detection and Cell Membership Specification by Principal Component Analysis

Abstract

In 1949, Donald Hebb postulated that assemblies of synchronously activated neurons are the elementary units of information processing in the brain. Despite being one of the most influential theories in neuroscience, Hebb's cell assembly hypothesis only started to become testable in the past two decades due to technological advances. However, while the technology for the simultaneous recording of large neuronal populations undergoes fast development, there is still a paucity of analytical methods that can properly detect and track the activity of cell assemblies. Here we describe a principal component-based method that is able to (1) identify all cell assemblies present in the neuronal population investigated, (2) determine the number of neurons involved in ensemble activity, (3) specify the precise identity of the neurons pertaining to each cell assembly, and (4) unravel the time course of the individual activity of multiple assemblies. Application of the method to multielectrode recordings of awake and behaving rats revealed that assemblies detected in the cerebral cortex and hippocampus typically contain overlapping neurons. The results indicate that the PCA method presented here is able to properly detect, track and specify neuronal assemblies, irrespective of overlapping membership.

Detecting cell assemblies in large neuronal populations

“…Lopes-dos- Santos et al, 2011;Peyrache et al, 2010), independent component analysis (Laubach et al, 1999), shuffling methods for detecting repeated firing sequences (Abeles and Gat, 2001;Abeles and Gerstein, 1988;Berger et al, 2010;Gansel and Singer, 2012), and methods based on information theory (Arabzadeh et al, 2004;Quiroga and Panzeri, 2009), among others. In this review we focus on linear methods for detecting and tracking the activity of cell assemblies embedded in large neuronal populations.…”

“…Regarding step 2, here we compare surrogate methods with an analytical threshold recently introduced in Peyrache et al (2009), and, regarding step 3, we compare previously published methods (Chapin and Nicolelis, 1999;Laubach et al, 1999;Lopes-dos-Santos et al, 2011;Peyrache et al, 2010) and suggest improvements.…”

“…Although this theoretical bound was derived for large matrices, it is also a robust threshold for smaller matrices (Lopes-dos-Santos et al, 2011;Plerou et al, 2002). Nevertheless, Peyrache et al (2010) proposed the use of a finite sample bias correction based on the Tracy-Widom distribution (Tracy and Widom, 1994).…”

“…Altogether, the results indicate that the theoretical bound derived from the Marčenko-Pastur distribution is well suited for tracking the number of cell assemblies, since surrogate methods give similar results but are computationally demanding. More detailed parametrical studies on the Marčenko-Pastur distribution can be found in Lopesdos- Santos et al (2011). It is nevertheless advisable to compare the analytical and shuffling methods when working with actual spike recordings, since the specific statistical properties of a given dataset may be different from the simulated examples employed here.…”

“…4C bottom panels. Of note, since we are concerned with co-activation patterns, the main diagonal of P is set to zero, which assures that isolated activations of assembly members do not contribute to R (Lopes-dos- Santos et al, 2011;Peyrache et al, 2010). Fig.…”