Fig. 3: Identification of a smooth, monotonic relationship between E/I-ratio and FC to fit brain network models.
From: Learning how network structure shapes decision-making for bio-inspired computing
a Tuning curves for a reduced model with only two nodes, but otherwise identical to the 379-nodes BNM. FC (that is, correlation) between the two nodes increased smoothly and monotonically as a function of their E/I-ratio \(\frac{{w}_{{{{{\mathrm{1,2}}}}}}^{{LRE}}}{{w}_{{{{{\mathrm{1,2}}}}}}^{{FFI}}}\). The relationship between E/I-ratio and FC persisted when the strength of noise \(\sigma\) (upper panel; Eqs. 5 and 6) and the strength of structural coupling \({C}_{{ij}}\) (lower panel; Eqs. 1 and 2) were modulated for test purposes (both are fixed parameters during the fitting of the full 379-nodes model). b Fitting results for the full 379-nodes model for one exemplary FC. Empirical (upper triangular portion of the matrix) versus simulated (lower triangular portion of the matrix) FC and joint distributions without E/I-tuning (upper panel) and with E/I-tuning (lower panel). c Pearson correlations and root-mean-square errors between all N = 650 empirical and simulated FCs for three different model variants: EI-tuning (the tuning algorithm applied on both \({w}_{{ij}}^{{LRE}}\) and \({w}_{{ij}}^{{FFI}}\)), E-tuning (the tuning algorithm applied only on \({w}_{{ij}}^{{LRE}}\)), original (tuning of a scalar global coupling scaling factor to rescale \({C}_{{ij}}\)).
