Fig. 2

Comparative sensitivity and sources of bias in randomization-based versus binomial calculation of network enrichment. Network enrichment between all vs. all 330 gene sets was analyzed with both NRZ and CSB methods. P-value distributions were compared using Q-Q plots (columns 1 and 2) and scatter plots of log (p) values (column 3). Q-Q plots in column 1 display both the total distributions (black lines), i.e. regardless of GS size, and distribution fractions that correspond to smaller GS (N _AGS + N _FGS, color lines). The QQ-plots in column 2 are insets of the black, un-stratified Q-Q plots of column 1. Identity lines (x = y) are plotted in dotted grey and dotted red in the QQ-plots and scatter plots, respectively. Analogous plots in regard of other factors biasing NRZ p-values (C _AGS + C _FGS and N _edges) are provided in Additional file 1: Figure S1 (note that plot columns 2 and 3 are the same as in the present figure)