Fig. 5: O-Information, brain topography, and functional specialization of optimally synergistic subsets identified by simulated annealing All panels show data from the HCP sample.

a Annealing was carried out 5000 times for each subset size. This plot shows O-information for each optimized subset (gray dots) and their mean (blue line). Note that annealing fails to converge onto any synergistic subsets for subsets containing more than 24 nodes. Optimally negative O-information is achieved for subsets between 8 and 12 nodes. For each subset of size k each node was removed individually and the O-information of the remaining k − 1 nodes was calculated. If the O-information was lower than given by the original k node subset, the contribution of that node was considered redundant and the synergy in the subset attributable to its k − 1 node counterpart. For the vast majority of subsets smaller than 12 nodes, no nodes could be removed in a way that increased synergy, indicating that these subsets consisted of nodes with irreducible synergy (red in Fig. 5a). b Frequency of individual node participation across optimally synergistic 10-node subsets (4021 unique subsets out of 5,000 annealing runs), displayed on a surface rendering of the cerebral cortex (cf. Fig. 4b). c Mean O-information (left) and fraction of synergistic subsets (right) encountered in samples of 20,000 subsets that contained nodes belonging to between 1 and 7 canonical FC systems (HCP data). The mean O-information for samples obtained exclusively from each of the 7 FC systems is indicated (red dots).