Anatomical-connectivity-guided functional connectivity reveals task-relevant pathways during proactive task-switching via recurrent graph neural networks.

In an increasingly complex society, cognitive flexibility is essential for effective multitasking. We developed a novel approach—Anatomical-Connectivity-Guided Functional Connectivity (ACG-FC)—and applied it to the preparatory phase of a task-cueing paradigm to examine functional coordination mechanisms among brain regions critical for cognitive flexibility during proactive control and temporal expectation. This approach emphasizes the importance of inter-regional collaboration by integrating the cortical structural connectome, which reflects transmission pathways with electroencephalography (EEG) data within recurrent graph neural networks (RGNNs) to evaluate the functional relevance of brain regions and their associated neural connections. In generalizing signal pattern differences across proactive control processes, RGNNs outperformed other symbolic machine learning and deep neural network (DNN) models commonly applied to neurophysiological data, indicating that spatial information from the structural connectome enhances EEG pattern recognition during executive control. Meanwhile, single-population attribution analysis revealed that general proactive task preparation involves functional coordination within three synchronous ACG-FC subnetworks clustered around the frontoparietal "multiple demand" system. This supports multiple late-latency neural priming processes during execution preparation, associated with functionally distinct yet highly complementary subnetworks in the prefrontal, cingulate, and temporal cortices. Furthermore, dual-population attribution analysis indicated that scattered ACG-FC differences specifically underlie distinct modes of proactive control or target anticipation. ACG-FC extends conventional functional connectivity (FC) from signal-dependent pairwise analyses to signal-structure-dependent network analyses by using a group-level anatomical connectome as a structural prior to guide the extraction of functional interactions, offering a new perspective on task relevance between cognitive activities and their neural bases.