Table 1. Performance of PCgen and residuals-based approaches, averaged over 500 simulated datasets per scenario.

	${\mathcal{G}}_{\mathcal{Y}}$				$G\to Y_j$
	TPR	FPR	TDR	SHD	TPR	FPR	TDR
Scenario 1 $\left(p=4\right)$	$\left(p_t=1/3\right)$				$\left(p_g=0.5\right)$
PCgen	0.647	0.006	0.981	0.442	0.982	0.026	0.995
PCres (replicates)	0.650	0.039	0.908	1.410
PCres (means)	0.521	0.390	0.438	3.174
Scenario 2 $\left(p=4\right)$	$\left(p_t=0.5\right)$				$\left(p_g=0.5\right)$
PCgen	0.804	0.033	0.986	1.246	0.976	0.031	0.995
PCres (replicates)	0.819	0.073	0.939	2.320
PCres (means)	0.672	0.364	0.659	3.628
Scenario 3 $\left(p=20\right)$	$\left(p_t=0.1\right)$				$\left(p_g=0.3\right)$
PCgen	0.895	0.002	0.985	6.806	0.969	0.018	0.991
PCres (replicates)	0.911	0.004	0.961	9.874
Scenario 4 $\left(p=100\right)$	$\left(p_t=0.01\right)$				$\left(p_g=0.1\right)$
PCgen	0.959	0.001	0.942	27.288	0.976	0.022	0.943
PCres (replicates)	0.962	0.001	0.940	38.410

SE for the TPR, FPR, and TDR were between 0.001 and 0.015. SE for the SHD were ∼0.06 (scenarios 1 and 2), 0.18 (scenario 3), and 0.28 (scenario 4). For the performance of other variants of PCgen and PCres in scenarios 1 and 2, see Table S2. In scenario 4, we used PCgen with the RG-test (PCgen-RG-uni); in the other scenarios we used the RC-test, with prior screening (PCgen-RC-screening). All acronyms are explained in Table 1 in File S2. PCres (replicates) and PCres (means) refer to PCres-uni-R and PC-multi-A.