Second-generation PLINK: rising to the challenge of larger and richer datasets

GigaScience

Table 6 Association analysis max(T) permutation test times (sec)

Other parameter(s)	Dataset	Machine	PLINK 1.07	PERMORY 1.1	PLINK 1.90	Ratio
–trend (C/C)	synth1	Mac-2	∼20 k		18.7	1.1 k
		Mac-12	∼16 k		2.8	5.7 k
		Linux32-2	∼21 k		65.0	320
		Linux64-512	∼17 k	285.0	2.8
		Win32-2	∼35 k	1444.2	61.5
		Win64-2	∼25 k	889.7	14.4
	synth2	Mac-2	∼170 k		42.4	4.0 k
		Mac-12	∼180 k		6.4	28 k
		Linux32-2	∼410 k		391.0	1.0 k
		Linux64-512	∼200 k	580.9	13.7
		Win32-2	∼1100 k	2362.5	198.0
		Win64-2	∼370 k	1423.6	34.0
–fisher (C/C)	synth1	Mac-2	∼150 k		21.9	6.9 k
		Mac-12	∼150 k		3.7	41 k
		Linux32-2	∼170 k		57.8	2.9 k
		Linux64-512	∼120 k		3.4	35 k
		Win32-2	∼440 k		64.9	6.8 k
		Win64-2	∼200 k		22.0	9.1 k
	synth2	Mac-2	∼890 k		49.8	18 k
		Mac-12	∼690 k		7.6	91 k
		Linux32-2	∼1300 k		393.7	3.3 k
		Linux64-512	∼720 k		13.0	55 k
		Win32-2	∼3600 k		208.3	17 k
		Win64-2	∼1700 k		35.6	48 k
–assoc (QT)	synth1	Mac-2	∼30 k		148.0	200
		Mac-12	∼22 k		22.6	970
		Linux32-2	∼68 k		847.2	80
		Linux64-512	∼29 k		29.2	990
		Win32-2	∼58 k		896.1	65
		Win64-2	∼36 k		264.2	140
–assoc lin (QT)	synth1	Mac-2			606.8
		Mac-12			34.7
		Linux32-2			3212.6
		Linux64-512		1259.8	46.4	27.2
		Win32-2		2115.7	3062.7	0.69
		Win64-2		972.6	336.6	2.89

All runs are with 10000 permutations and –seed 1. The PLINK 1.9 implementation extends Pahl et al.’s PERMORY algorithm [29] with multithreading (note the 12- and 512-core machine results) and additional low-level optimizations.

ISSN: 2047-217X