Probing the Functional Circuitry Underlying Auditory Attention via Dynamic Granger Causality Analysis

We consider the problem of inferring time-varying Granger causal interactions among multiple simultaneously recorded spike trains from a neuronal ensemble. We present a dynamic Granger causality measure with sparsity and adaptivity features for point process observations, and estimate it recursively. We develop a statistical inference framework based on asymptotic analysis of deviance, and perform a multiple-hypothesis testing procedure to control the false discovery rate. We apply our method to electrophysiology data recorded from the ferrets auditory and prefrontal cortices during a series of auditory tasks in order to probe the functional circuitry underlying auditory attention at the neuronal scale.