Multiple Granger causality tests for network structure estimation from time-series data

To identify network structures is a key for elucidating functions of various kinds of networks such as cortical local circuits. Granger causality (GC) test has been used for estimating directed network structure from time-course of neuronal activities. Although GC statistic for a pair of nodes can be substantially influenced by other nodes, ignoring such influence can degrade detection performance of multiple GC tests. To improve the multiple GC tests, and hence the estimation of large network structures, therefore, we propose an extension of GC by introducing optimal discovery procedure (ODP) that shows the best detection power in general multiple testing problems. Applying our proposed method to a benchmark dataset, we show the performance of estimating the network structure is improved over those by the exiting methods.