A method for the time-frequency analysis of high-order interactions in non-stationary physiological networks

<i>Objective</i>. Several data-driven approaches based on information theory have been proposed for analyzing high-order interactions (HOIs) involving three or more components of a network system. The existing methods do not account for temporal correlations in the data, or are defined only in the time domain and rely on the assumption of stationarity in the underlying dynamics, making them inherently unable to detect frequency-specific behaviors and track transient functional links in physiological networks.<i>Approach</i>. This study introduces a new framework which enables the time-varying and time-frequency analysis of HOIs in networks of random processes through the spectral representation of vector autoregressive models. The time- and frequency-resolved analysis of synergistic and redundant interactions among groups of processes is ensured by a robust identification procedure based on a recursive least squares estimator with a forgetting factor.<i>Main results</i>. Validation on simulated networks illustrates how the time-frequency analysis is able to highlight transient synergistic behaviors emerging in specific frequency bands which cannot be detected by time-domain stationary analyzes. The application on brain evoked potentials in rats elicits the presence of redundant information timed with whisker stimulation and mostly occurring in the contralateral hemisphere. The application to cardiovascular oscillations reveals a reduction in redundant information following head-up tilt, reflecting a functional disconnection within the physiological network of heart period, respiratory, and arterial pressure signals.<i>Significance</i>. The proposed framework enables a comprehensive time-varying and time-frequency analysis of the hierarchical organization of dynamic networks. As our approach goes beyond pairwise interactions, it is well suited for the study of transient high-order behaviors arising during state transitions in many network systems commonly studied in physiology, neuroscience and other fields.