Fig. 2

Comparison of PUC and FDR. a Two regulatory networks are simulated independently, then both networks’ node expression levels combined into one data set. In a correlation network constructed from the simulated data, any correlations (links) between nodes from independent networks are known to be erroneous; Bayesian simulations (b), as well as gene regulatory simulations performed GeneNetWeaver (c) suggest that PUC more accurately reflects network error than FDR (Benjamini-Hochberg, FDR-BH); as network size grows, PUC more accurately reflects network error than FDR-BH (d) or its variation with multiple hypothesis under dependence called FDR Benjamini-Yekutieli (FDR-BY) (e); PUC correlates with FDR in both gene expression (f) and macroeconomic (g) data