Table 7.

Empirical and simulation results for BFDP procedure

Trait	Threshold (Bayesian FDR)	Empirical						Simulation
		$\widehat{{\pi }_1}$ a	Positives		eFDRc	$\widehat{{\pi }_1}$ d	Positives			eFDR
			False	Trueb			False		True
HDL (ntest = 19,840 ntruth = 188,577)	5%	1.3 × 10−4	1	17	5.6%	8.7 × 10−5	0.41		10	4.0%
	10%		4	17	19%		0.76		12	6.1%
	15%		6	18	25%		1.2		13	8.4%
LDL (ntest = 19,840 ntruth = 188,577)	5%	1.3 × 10−4	2	17	11%	9.6 × 10−5	0.37		14	2.5%
	10%		5	17	23%		0.83		16	4.9%
	15%		6	18	25%		1.3		18	7.0%
TG (ntest = 19,840 ntruth = 188,577)	5%	2.1 × 10−4	1	10	9.1%	1.6 × 10−4	0.36		9.8	3.6%
	10%		4	10	29%		1.0		11	8.4%
	15%		4	12	25%		1.6		12	12%
Height (ntest = 133,653 ntruth = 693,529)	5%	2.0 × 10−3	2	338	0.59%	2.9 × 10−3	28		317	8.1%
	10%		7	406	1.7%		51		356	13%
	15%		9	468	1.9%		76		385	17%
BMI (ntest = 123,865 ntruth = 681,275)	5%	3.6 × 10−4	0	35	0%	5.2 × 10−4	3.9		67	5.4%
	10%		0	43	0%		7.2		82	8.0%
	15%		0	50	0%		11		93	10%

^a $\widehat{{\pi }_1}$ is the estimated prior probability of association at a variant site equal to the proportion of tested variants with P < 5 × 10⁻⁸.

^bNumber of loci in truth set for HDL: 89, LDL: 72, TG: 60, height: 1100, BMI: 724.

^ceFDR is calculated as number of false positives divided by sum of true and false positives.

^dAverage $\widehat{{\pi }_1}$ in 1,000 replicate datasets.