Figure 6.

Estimated critical boundaries for an ensemble of empirical models, using an alternative averaging scheme. The leftmost panel represents the traditional in-degree approach, while the central and rightmost ones consider our proposed measures. The activation spread parameters ${\left({\sigma }^2\right)}^{\prime }$ and $H^{\prime }$ are computed by averaging node activation variance and entropy, respectively. The color of each point indicates the network’s Derrida coefficient. The curves are estimates of the critical boundary obtained from the optimal boundary in the Cell Collective data, which simultaneously maximizes the MCC, accuracy, and Cohen kappa metric in all three cases.