Macroscopic resting-state brain dynamics are best described by linear models.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Large neural networks are often assumed to exhibit complex nonlinear behaviors.
• This assumption is prevalent in understanding brain function and dynamics.

Purpose of the Study:

• To challenge the assumption of widespread nonlinear dynamics in large neural networks.
• To investigate the suitability of linear versus nonlinear models for describing macroscopic brain activity.

Main Methods:

• Utilized mathematical models based on intracranial electroencephalography (iEEG) local field potentials and functional magnetic resonance imaging (fMRI) blood-oxygen-level-dependent (BOLD) signals.
• Applied state-of-the-art linear and nonlinear model families to resting-state brain activity data from Human Connectome Project (N=700) and Restoring Active Memory project (N=122).
• Evaluated models based on predictive power, computational complexity, and unexplained residual dynamics.

Main Results:

• Linear autoregressive models demonstrated the best fit across both iEEG and fMRI data types and all performance metrics.
• Microscopic nonlinear dynamics appear to be masked by macroscopic factors such as spatial/temporal averaging, observation noise, and limited data.
• These factors, inherent to aggregated brain activity and technological constraints, obscure underlying nonlinearities.

Conclusions:

• Linear models can faithfully describe macroscopic brain dynamics during resting-state conditions.
• The apparent complexity of neural activity may be an artifact of measurement and data aggregation rather than true microscopic nonlinearity.
• Re-evaluation of the necessity for complex nonlinear models in macroscopic brain dynamics analysis is warranted.