FCCA: A New Method of Constructing Causality Network Based on Graph Structure Information and Conditional Causality Test

Pairwise granger causality test, which detects the causal connectivity between two nodes in a graph, has been widely used in various fields since it was proposed by economist Granger in 1969. However, pairwise granger causality test has the drawback of generating false positive causality, which is an indirect causal influence between two nodes mediated through a third node. In 1984, Geweke proposed the conditional Granger causality model, which enabled the model to eliminate false positive causal connectivity and accurately identify the causal relationships between two nodes in a high-dimensional dataset. The Matlab software tool GCCA realizes the calculation of conditional causality. For a given network, GCCA finds out all the triangular causal relationships (X~Y, Y~Z, X~Z) and calculates the causality among all three nodes. However, it is not necessary to calculate among all the three-node combinations as there may not be significant causal connectivity between any given two nodes. In additional, the full calculation of conditional granger causality could be slow. Here, we proposed a new test named Fast Causal Connectivity Analysis (FCCA) as a fast and approximative test for causal connectivity. We compared the performance of GCCA and FCCA using a time series fMRI dataset and showed that FCCA has acceptable accuracy and theoretically faster run time.