FIG 4 .

Comparison of FDR methods on data-driven simulations with signals. (a) Number of hypotheses after filtering in the DIBD simulation, where the initial (nonfiltered) number of hypotheses is 867, for 15 samples per group and 10% true signals. (b) Number of hypotheses after filtering in the CS simulation, where the initial (nonfiltered) number of hypotheses is 2,817, for 15 samples per group and 10% true signals. (c) Estimated FDR for the DS, FBH, and BH procedures (red, orange, and blue lines, respectively) as a function of filter level in the DIBD simulation (15 samples per group) in panel a. The green line indicates the nominal FDR control level (0.1). (d) Same as panel c for the CS simulation. (e) Number of truly differential OTU discoveries for the DS, FBH and BH procedures (red, orange, blue, respectively) as a function of filter level in panel a. (f) Same as panel e for the CS simulation. (g) Estimated FDR for the DS, FBH and BH procedures (red, orange, blue, respectively) as a function of number of samples in each group in the DIBD simulation (15 samples per group, filter level at 10). Green line indicates the nominal FDR control level (0.1). (h) Same as panel g for the CS simulation. (i) Number of truly differential OTU discoveries for the DS, FBH, and BH procedures (red, orange, and blue lines, respectively) as a function of number of samples in each group in panel g. (j) Same as panel i for the CS simulation.