A Whole-Mouse-Brain Model of Coupled Neural Masses: Analysis of Network Dynamics

Abstract

Neural mass models coupled through empirical structural connectomes are widely used to simulate whole-brain activity, but the dynamics of the resulting large, delay-coupled networks are rarely analyzed directly. We present a dynamical systems analysis of 434 Jansen-Rit neural masses coupled by the Allen Mouse Brain Connectivity Atlas at the canonical single-node parameter values. Pseudo-arclength continuation of the network equilibria in the global coupling scale and in the baseline external input, with stability determined from the characteristic roots of the delayed linearization by Beyn's contour-integral method, indicates where the network must oscillate and where it can rest. Simulations show that the oscillatory regime is chaotic: the maximum Lyapunov exponent is positive for the empirical connectome, grows with network size and coupling scale, and is insensitive to the conduction delays, but it is suppressed when the arrangement of connection weights is flattened, shuffled, or rewired. The same weight modifications that suppress chaos drive the network toward full synchrony and abolish metastability, as measured by the Kuramoto order parameter. We discuss the consequences of these findings for fitting whole-brain neural mass models to empirical data.