Graph embedding of structural connectivity characterizes task‐evoked brain dynamics

Abstract

Task‐evoked brain dynamics are sensitive to dysfunction but are often difficult to acquire, and whether more readily obtainable features, such as structural features, can provide an informative characterization of abnormal task‐evoked functional communication remains unclear. To address this question, we analyzed multimodal MRI data from 39 patients with schizophrenia and 67 normal controls from the UCLA Consortium for Neuropsychiatric Phenomics dataset. We derived high‐order structural embedded features from diffusion‐weighted imaging using graph embedding methods and quantified functional communication from resting‐state and task‐based functional magnetic resonance imaging using activity‐flow information transfer mapping. Linear regression was used to assess the relationship between structural embeddings and information transfer patterns. We further examined structural and functional alterations in schizophrenia during an episodic memory task and performed simulation‐based perturbation analyses to evaluate whether virtual strengthening of structurally abnormal connections was associated with shifts in aberrant brain activity toward the control pattern. The results showed that structural embeddings were more strongly associated with information transfer than conventional structural connectivity or resting‐state functional connectivity. In addition, the accuracy of structural embedding‐based prediction of information transfer was associated with clinical symptom severity. Simulation analyses further indicated that enhancing structurally abnormal connections was associated with improved information transfer prediction and partial shifts in functional activation toward the control pattern. These findings support graph‐embedded structural connectivity as a candidate connectome‐derived imaging biomarker of abnormal task‐evoked brain dynamics.