A model of the mouse cortex with attractor dynamics explains the structure and emergence of rsfMRI co-activation patterns

Abstract

Neural network models have been instrumental in revealing the foundational principles of whole-brain dynamics. Here we describe a new whole-cortex model of mouse resting-state fMRI (rsfMRI) activity. Our model implements neural input-output nonlinearities and excitatory-inhibitory interactions within areas, as well as a directed connectome obtained with viral tracing to model interareal connections. Our model makes novel predictions about the dynamic organization of rsfMRI activity on a fast scale of seconds, and explains its relationship with the underlying axonal connectivity. Specifically, the simulated rsfMRI activity exhibits rich attractor dynamics, with multiple stationary and oscillatory attractors. Guided by these theoretical predictions, we find that empirical mouse rsfMRI time series exhibit analogous signatures of attractor dynamics, and that model attractors recapitulate the topographical organization and temporal structure of empirical rsfMRI co-activation patterns (CAPs). The richness and complexity of attractor dynamics, as well as its ability to explain CAPs, are lost when the directionality of underlying axonal connectivity is neglected. Finally, complexity of fast dynamics on the scale of seconds was maximal for the values of inter-hemispheric axonal connectivity strength and of inter-areal connectivity sparsity measured in real anatomical mouse data.