Estimating Functional Connectivity in Networks of Dissociated Cortical Neurons

Abstract

Event Abstract Back to Event Estimating Functional Connectivity in Networks of Dissociated Cortical Neurons Matteo Garofalo1*, Alessandro N. Ide2, Thierry Nieus1, Michela Chiappalone1 and Sergio Martinoia2 1 Italian Institute of Technology, Italy 2 University of Genova, Italy Background: Identification of causal relationships between pairs of neurons plays an important role in the study of synaptic interactions at population level. To investigate functional connectivity in cortical networks we used cross-correlation and information theory based methods. Methods: Dissociated cortical neurons were obtained from rat embryos and plated onto microelectrode arrays (MEAs). Neural networks based on Izhikevich cell models (Izhikevich, 2003), able to reproduce MEA experimental data, was used. Functional connectivity for each couple of channels or neurons was estimated by using (i) cross-correlation, (ii) mutual information (MI) and (iii) entropy. (i) The strength of the connection between two considered channels was evaluated on the basis of the value of the cross-correlogram peak, which had to be above a pre defined threshold. The peak latency was also evaluated in order to determine the direction of the connection. (ii) MI was evaluated to account for the spike count and time code contribution between each pair of channels. Spike trains were shifted and a MI function was defined in order to recover the connections directionality. (iii) Joint inter spike interval (J-ISI) entropy was evaluated to assess causality between two channels. All the methods were validated on neural networks characterized by different connectivities. Performances were measured in terms of ROC (receiver operating characteristic), in particular by using the AUC parameter (area under ROC) (Fawcett, 2006). Results: MI proved to detect efficiently the connectivity maps in networks characterized by bidirectional connections. Since maps inferred by methods based on MI are symmetrical, in more realistic networks with many unidirectional connections its performances break down. MI functions permit also to recover the connections' directionality. Average ROCs (Figure1), computed on nine different neural network models, are shown. The joint entropy method displays (Figure1) higher performances (AUC=0.94) compared to the correlation method (AUC=0.87). Nevertheless the performances are comparable, the differences in the connectivity maps inferred using method (i) and (iii) on real data are evident. Conclusions: Entropy and cross-correlation have shown good results in terms of AUC and ROC in model data. However, they predict different connectivity maps on real data. Despite MI can not recover unidirectional connections it correlates better, compared to entropy and correlation, with the synaptic strength. Preliminary results obtained through MI functions have shown the capacity to recover the links directionality and, therefore, we are investigating the connectivity maps that methods (ii) predicts. The aim is to combine the advantages of each method to improve the global performances of our approach. 07-28-2008-15-04_Entropy Entropy tn_Entropy tn_07-28-2008-15-04_Entropy