Fig. 1: Schematic of parametric mean-field model (pMFM) optimization.
From: Sensory-motor cortices shape functional connectivity dynamics in the human brain
A The pMFM comprised ordinary differential equations (ODEs) at each cortical region coupled by a structural connectivity (SC) matrix. The circuit-level parameters were allowed to vary across cortical regions, parameterized by a linear combination of resting-state functional connectivity (FC) gradient and T1w/T2w spatial maps. The pMFM was used to generate simulated static FC and functional connectivity dynamics (FCD). The Covariance Matrix Adaptation Evolution Strategy (CMA-ES) was used to estimate the pMFM by minimizing a cost function of disagreement with empirically observed FC and FCD. B The CMA-ES algorithm was applied to the Human Connectome Project (HCP) training set (N = 351) to generate 5000 candidate parameter sets. The top 10 candidate parameter sets were then selected from the 5000 candidate sets based on the model fit in the validation set (N = 350). Finally, these top 10 candidate sets were evaluated in the HCP test set (N = 351). Comparison of the pMFM with other parametrizations (Fig. 3 and Supplementary Fig. S3) utilized the same training-validation-test procedure.
