Independence-based causal discovery analysis reveals statistically non-significant regions to be functionally significant
Background and Hypothesis: Traditional fMRI analyses often ignore regions that fail to reach statistical significance, assuming they are biologically unimportant. We tested the accuracy of this assumption using causal discovery based-analysis that go beyond associations/correlations to test the causality of influence of one region over the other. We hypothesized that the network of statistically significant (active network, AN) and non-significant regions (silent network, SN) causally interact and their features will causally influence psychopathology severity and working memory performance. Study Design: We examined AN and SN during N-BACK task on 25 FHR and 37 controls. Clusters with significantly different activations were juxtaposed to 360 Glasser atlas parcellations. The PC algorithm for causal discovery was implemented. Connectivity of regions with the highest alpha-centrality (HAC) were examined. Results: Seventy-seven Glasser regions were in the AN and the rest were silent nodes. Two regions showed HAC for FHR and HC. Among controls, one HAC region was silent (auditory association cortex) and the other one was active (insula). Among FHR, both were silent nodes (early auditory cortex). These HAC regions in both groups had bidirectional directed edges between each other forming a reciprocal circuit whose edge-weights causally increased magical ideation severity. Conclusion: Causal connectivity between SN and AN suggests that the statistically non-significant and significant regions influence each other. Our findings question the merit of ignoring statistically non-significant regions and exclusively including statistically significant regions in the pathophysiological models. Our study suggests that causality analysis should receive greater attention.