Description length guided nonlinear unified Granger causality analysis

Abstract Most Granger causality analysis (GCA) methods still remain a two-stage scheme guided by different mathematical theories; both can actually be viewed as the same generalized model selection issues. Adhering to Occam's razor, we present unified GCA (uGCA) based on minimum description length principle. In this research, considering common existence of nonlinearity in functional brain networks, incorporated nonlinear modeling procedure into proposed uGCA method, which an approximate representation Taylor's expansion was adopted. Through synthetic data experiments, revealed that obviously superior its linear and conventional GCA. Meanwhile, characteristics high-order terms cross-terms would successively drowned out noise levels increased. Then, real fMRI involving mental arithmetic tasks, further illustrated these may indeed at high level, hence causal sufficient. Next, autism spectrum disorder patients data, compared with GCA, network property connections obtained appeared more consistent clinical symptoms.