Multivariate partial coherence analysis for identification of neuronal connectivity from multiple electrode array recordings

Studying the neuronal pattern of interactions may help us to understand the underlying processes of functional connectivity in the brain. Simultaneous recording of multiple neuronal activities using a multi electrode array provides rich neuronal signals and requires appropriate statistical and computational methods to demonstrate any connectivity. Using ordinary coherence to infer true connectivity between two neurons is subject to influence from other intermediate neurons (predictors). This study uses multivariate partial coherence analysis to estimate the synchronization of spiking neurons from recorded signals. The objective is to develop an algorithm to differentiate between conditional and unconditional independence among neurons. This may provide useful information on how neurons interact to transmit or receive signals by taking into account the influence from the predictors. In this paper, we validate the method of multivariate partial coherence analysis on a network of spiking neurons consisting of nine interconnected neurons simulated by the Izhikevich model. Then we implement this method on physiological signals to infer the connectivity among ten neurons recorded across different areas of rat hippocampus. Our analyses show the applicability of the proposed method for identifying the true connectivity in the simulated data. We also present the effect of predictor neurons on pairwise indirect relationships within specific frequency content. Conditional independence links in real data are reduced by 53% compared with unconditional links. The proposed method could be a valuable tool for observing the connectivity between neurons during normal and abnormal cognitive responses.