Fig. 5

Comparing DGCA to alternatives segregated by the strength of correlation difference. a-c: Representative receiver operating characteristic (ROC) curves show the ability of DGCA a, EBcoexpress b, and Discordant (with the use of the Fisher z-transformation) c to accurately detect truly differential correlated gene pairs in each of the six differential correlation classes that specify a difference in correlations between conditions, as well as when comparing all gene pairs with a gain in correlation (GOC) or a loss of correlation (LOC) in one condition compared to the other, at a sample size of n = 30. d-e: Comparison of area under curve (AUC) statistics for 5 runs of the simulation study using each of the three methods at different numbers of samples, where errors bars represent the standard error of the mean, segregated to only those gene pairs with a strong difference in correlation (absolute value of difference in ρ = 1; d) or segregated to only those gene pairs with a medium strength difference in correlation (absolute value of difference in ρ = 0.5; e). Asterisks indicate a Bonferroni-adjusted significant difference in the AUCs (p < 0.0083) between DGCA and each of the other methods, tested using a two-sided t-test