Figure 3.

Average GO-distances for biological process, molecular function and cellular component for the regulatory network of S. Cerevisiae (black curve), and its randomized counterparts. As ε → 0 one increase the bias for generating random networks with maximal proximity (similarity) of the GO-annotations of neighboring proteins. Left column (a,c,e) analyze the direct GO-distances, whereas the right column (b,d,f) analyze the hierarchical GO-distances as function of distance in the real and the randomized regulatory networks. In (g) is shown how we randomize the networks with probability ε : A random regulatory network is generated from the real one by multiple rewirings of pairs of regulatory links. For each rewiring one select two random connections A → B, X → Y and consider rewiring to a network where instead A → Y and X → B. With probability ε one always rewire. With probability 1 - ε one finds a random pair of links where the GO-distances after are smaller or equal to the GO-distances before the swap. That is, D(A,Y) ≤ D(A,B) and D(X,B) ≤ D(X,Y), here symbolized by nodes of similar colors being brought closer together.