Irene Malvestio —  Università degli Studi di Firenze - CNR Firenze - Universitat Pompeu Fabra Barcelona #
Nonlinear interdependence detection from spike trains #
The detection of interdependence of unknown dynamics from their signals is an important problem in many different fields. In particular in neuroscience, assessing the connectivity between neurons is a crucial task in order to understand the architecture of the brain. Initially, continuous signals like EEG and MEG were studied to reconstruct connectivity on a large scale. More recently the focus of attention has shifted to the analysis of microscopic data recorded from individual neurons in the form of discrete spike trains.
Here we describe an approach based on the asymmetric state similarity criterion, in the formulation of the interdependence measure L [1]. It is an extension of a method for continuous signals [2], and it can detect both linear and nonlinear coupling. The approach is modular, for example different spike train distances can be used to assess similarity [3]. With tests on the Hindmarsh-Rose model system we show that this method is robust to noise and versatile with respect to different spike train regimes. Furthermore, its modularity leads to sensitivity to different coupling intensities. In closing we discuss the necessity of surrogate techniques for assessing significance in the application to real data. The implementations of the measure L and of the spike train distances are available online [4].

References: 
[1] R.G. Andrzejak, T. Kreuz. EPL (Europhysics Letters) 96, 50012 (2011)
[2] D. Chicharro, R.G. Andrzejak. Physical Review E 80, 026217 (2009)
[3] T. Kreuz. Scholarpedia 6, 11934 (2011)
[4] http://ntsa.upf.edu/downloads; http://www.fi.isc.cnr.it/users/thomas.kreuz/sourcecode.html