A novel analysis framework for characterizing ensemble spike patterns using spike train clustering and information geometry.
1. NSMA, Univ Arizona, Tucson, AZ, USA
2. RIKEN BSI, Wako-shi, Saitama, Japan
Presently, hundreds of neurons can be recorded simultaneously from different brain areas of behaving animals. It is conjectured that multi-neuronal spike-synchronization patterns (‘assemblies’) emerge dynamically and may play an important role in cognitive functions. Their detection has proven difficult because they are hidden among the overall recorded neuronal groups. Furthermore, the quantification of their significance has also proven difficult because correlation due to spike-timing relations among neurons cannot be easily separated from correlation due to mean firing rate modulations of individual neurons.
We propose a novel analysis framework for the characterization of multi-neuronal spike patterns that resolves these difficulties by integrating two independently developed analysis methods based on 'spike train clustering' and 'information geometry'. With the former method, neuronal subgroups that exhibit synchrony are first identified. With the latter method, 'pure' correlations due to spike-timing relationships among neurons in those neuronal subgroups are separated from correlations due to mean firing rate modulations.
We examined the proposed method using ensemble spike trains that were generated by recurrent networks of biophysical model neurons connected by AMPA and GABAA synapses. Correlation was introduced by either common external inputs or by the modulation of specific intrinsic connections. The spike train clustering method identified subgroups of synchronized neurons successfully and dramatically reduced the number of neuron pairs that needed to be analyzed. Information geometry applied on these subgroups successfully detected pure, rate-independent correlations. The advantages of using information geometry over the conventional correlation measures such as the correlation coefficient were also examined. These results indicate that spike train clustering and information geometry are potentially powerful tools for the detection and analysis of multi-neuronal spike patterns.
Supported by MH046823, JST/JSPS Overseas Research Fellowship