Figure 5
Clustering of k-dependent and w-dependent EGs. (A) Relative frequency of GO terms enriched with k-dependent and w-dependent EGs. (B) Decrease of enriched GO terms due to removal of the same number of k-dependent or w-dependent EGs. (C) n-cliques and their biases toward k-dependent and w-dependent EGs. The normal distribution of z values was approximated by binomial trials with success probability given by the fraction of cliques biased toward k-dependent EGs. (D) Change in the link density of a community upon removal of a single node (ΔD) at different hierarchical levels (fH). Hierarchical community structure was searched using the Walktrap algorithm. The gray line indicates the average ΔD for non-EGs (area, standard error). (E) Difference in ΔD between k-dependent and w-dependent EGs in different PPI networks. (F) Examples illustrating EPC2, a w-dependent EG, and UBE2I, a k-dependent EG, for their impact on communities. The protein of interest (blue circles, EPC2; red circles, UBE2I) and its interactions (green lines) with its first neighbors (green circles) are shown in the community. Numbers in parentheses indicate the rank percentile for k and w.
