Table 1.

Benchmarks

Dataset	Running Time
GCTA	pylmm	ALBI (0.001)	ALBI (0.01)
GTEx: 100 random samples	1.05 min	3.94 s	0.27 s	0.04 s
LURIC: 100 random samples	3.98 min	5.7 s	1.27 s	0.12 s
NFBC: 100 random samples	>4 hr	18.03 s	2.79 s	0.34 s
GTEx: 1,000 random samples	4.68 min	37.9 s	1.9 s	0.27 s
LURIC: 1,000 random samples	54.6 min	54.15 s	8.85 s	0.73 s
NFBC: 1,000 random samples	>1 day	2.30 min	25.82 s	3.21 s

Running times of ALBI and other brute-force methods. We compare the computational costs of ALBI to those of pylmm⁴² and GCTA⁴¹ (see Material and Methods). We generated the estimator distribution for h² = 0.5. For ALBI, we estimated the distributions at a precision of 0.01 and 0.001 by using 100 and 1,000 random samples. For pylmm and GCTA, we generated 100 and 1,000 random phenotypes. GCTA did not converge for ∼7% of the random samples. Running times are reported for the GTEx (185 individuals), LURIC (867 individuals), and NFBC (2,520 individuals) datasets.