Figure 4

Cancer perturbations may increase the entropy rate on networks withscale-free topology but not on random Poisson graphs:

(A) A cartoon of the network perturbation analysis: each node i of thenetwork is perturbed in turn by changing its expression value. The case ofoverexpression is here indicated in red. The increased expression draws insignaling flux from neighbours (only one perturbed edge is shown). Theentropy rate of the network after perturbing node i,SR_i, is computed and compared to the entropy rateSR of the original unperturbed network. For n nodes in thenetwork we get a distribution of entropy rate changes (SR_i− SR, i = 1,…,n). (B) Zoomed-inversion of a network perturbation, whereby a node i undergoes aperturbation (here overexpression). From the perspective of a neighbouringnode j, the perturbation causes a low signaling entropy configurationaround node j. Key question is how does this perturbation affect theglobal entropy rate. (C) Perturbation analysis result, in which each node(gene) of the network was perturbed through overexpression (red) orunderexpression (green). Plotted is the global entropy rate (SR) after theperturbation (y-axis) against the degree of the perturbed node (x-axis), for3 different networks: Erdos-Renyi (ER) graph, scale-free (SF) network andthe full PPI network (PPI). Black dashed line denotes the entropy ratebefore the perturbation. In each plot there as many data points as there arenodes in the network, each value corresponding to the perturbation of onlyone node. Number of nodes (nn), average degree (avK) and median degree(medK) are given.