Table 5.

The performance of estimated FDR methods for window-based rare variant tests, using three different methods for estimation.

Analytic strategy	Identification of windows containing causal variants			Identification of windows either containing causal variants, or correlated with such windows at r ≥ 0.90
	Benjamini–Hochberg (BH)	Beta-uniform mixture (BUM)	FDRtool	Benjamini–Hochberg (BH)	Beta-uniform mixture (BUM)	FDRtool
N—All	0.981 (0.033)	0.982 (0.030)	0.982 (0.030)	0.935 (0.080)	0.925 (0.088)	0.936 (0.079)
N—Strat	0.980 (0.036)	0.980 (0.042)	0.980 (0.036)	0.952 (0.049)	0.944 (0.056)	0.953 (0.049)
P—All	0.974 (0.064)	0.970 (0.107)	0.975 (0.060)	0.925 (0.096)	0.911 (0.130)	0.927 (0.091)
P—Strat	0.971 (0.075)	0.970 (0.073)	0.972 (0.075)	0.943 (0.078)	0.932 (0.087)	0.943 (0.078)
S—All	0.927 (0.038)	0.927 (0.037)	0.920 (0.040)	0.874 (0.046)	0.858 (0.046)	0.862 (0.045)
S—Strat	0.927 (0.037)	0.926 (0.038)	0.920 (0.040)	0.882 (0.045)	0.868 (0.045)	0.872 (0.046)

Results show the proportion of windows containing no causal variants (tFDR) for an estimated FDR of 0.05. Results are shown for the simulation with σ = 0.5. Results are the mean across 100 simulations (standard deviation).