Partial directed coherence and neuronal connectivity inference

After a brief review of the newly introduced concept of partial directed coherence (PDC), we discuss its role and limitations in disclosing the connectivity of networks comprising spiking neurons. Because of the inherent point- process nature of the signals involved, the problem must first be formulated in continuous time and subsequently rephrased in discrete time for computationally efficient processing purposes. This procedure, which we term "signal reconstruction" involves convolving the impulses associated to neuronal discharges with suitably denned "kernels" , i. e., superposed continuous time waveforms that are then discretized in time. We compare three such kernel candidates and show, via simulations of interconned neurons (leaky- and integrate-and-fire units), that kernel duration has substantial impact on the observed attainable confidence levels of connectivity inference according to network specifics involving its dynamics and its topology.