A new Framework for the Spectral Information Decomposition of Multivariate Gaussian Processes

Area of Science:

• Dynamical Systems Analysis
• Information Theory
• Neuroscience

Background:

• Existing information-theoretic measures primarily operate in the time domain.
• Oscillatory content in physiological systems is typically analyzed in the frequency domain.
• Previous work established links between information and spectral measures for Gaussian systems but not for higher-order interactions.

Purpose of the Study:

• To introduce a frequency-domain information-theoretic framework for analyzing higher-order interactions in multivariate dynamical systems.
• To quantify shared information, redundancy, and synergy between a target process and multiple sources.
• To bridge the gap between time-domain information theory and frequency-domain spectral analysis.

Main Methods:

• Development of a novel information-theoretic framework operating in the frequency domain.
• Simulation of linear interacting processes to validate the framework.
• Application of the framework to electroencephalography (EEG) time series data from a motor execution task.

Main Results:

• The proposed framework successfully quantifies shared information in specific frequency bands, outperforming time-domain measures for certain interactions.
• Demonstrated capability to detect information sharing not discernible by traditional time-domain methods.
• Successfully applied to EEG data, highlighting its utility in analyzing brain activity.

Conclusions:

• The frequency-domain information-theoretic framework provides a powerful tool for analyzing complex interactions in dynamical systems.
• This approach enhances the understanding of information flow and organizational principles in both simulated and biological systems.
• The framework offers novel insights into brain activity during motor tasks by analyzing neural oscillations.