Fig. 4: A generative network model of neonatal networks.

a Energies from 90,000 simulations fit to the consensus network. The scatter plot contains the energies of the best 10 models for each ‘value’ rule. The average energy landscapes of the two best-performing ‘value’ rules, both of which are homophily principles, depict the energies for all simulations in the parameter space. Landscapes include a green point indicating the location of the best-fitting model. While ‘neighbors’ had the lowest overall energy, ‘matching’ had the lowest energy averaged over the top 10 models. b Representative well-fit observed network and simulated ‘matching’ model network. (c) The average energy landscape of the individually fit ‘matching’ models. The white dots indicate the location of the best-fit models for all participants. The violin plot demonstrates the distribution of energies of the best-fit models (min = 0.07, max = 0.22, M = 0.09, SD = 0.01). d The topological fingerprints of observed and simulated networks depict the average correlations between local organization measures. e Global organization of observed and simulated networks were significantly correlated (using Pearson’s correlation) for global efficiency (p = 6.58 x 10^-155), number of rich club connections (p = 6.98 x 10^-9), average length of rich club connections (p = 3.84 x 10^-7), characteristic path length (p = 3.76 ×10^-91), and modularity (p = 2.49 x 10^-15). f Degree displayed significant spatial embedding using LMEM (p = 4.96 x 10^-5). The surface plot depicts the average degree in the left hemisphere for the observed and simulated networks in all regions. The top right graph depicts correlational data between observed and simulated degree for every participant (grey) and the sample average (red). The bottom graphs show the cumulative density function of the average observed (purple) and simulated (orange) networks. Shaded area around best fit lines represent 95% confidence intervals. *** indicates p < 0.001, ** indicates p < 0.01, * indicates p < 0.05. The node plots in panel (b) and surface plots in panel (f) are adapted from AAL90 atlas in Shi F, et al. (2011) Infant Brain Atlases from Neonates to 1- and 2-Year-Olds. PLoS ONE 6(4): e18746. doi:10.1371/ journal.pone.0018746.