Detecting purely epistatic multi-locus interactions by an omnibus permutation test on ensembles of two-locus analyses

Table 1 Computational time required by 2LOmb, a combined 2LOmb and MDR approach, and direct MDR analysis to detect interactions in simulated data sets with different sizes and different numbers of causative SNPs.

		Computational Time Required by Each Approach (sec)
		2LOmb			2LOmb+MDR			MDR
Number of Causative SNPs	Sample Size	1,000 SNPs	2,000 SNPs	4,000 SNPs	1,000 SNPs	2,000 SNPs	4,000 SNPs	100 SNPs
2	400	15	37	135	17	39	137	7,656
	800	21	59	224	23	61	226	15,990
	1,600	36	106	400	38	108	402	31,222
3	400	22	43	140	24	45	142	7,721
	800	30	65	229	32	67	231	16,206
	1,600	50	115	406	52	117	408	31,232
4	400	32	55	150	34	57	152	7,841
	800	46	80	236	48	82	238	16,285
	1,600	70	133	419	72	135	421	31,637

Only one computing processor in a Beowulf cluster is occupied during the analysis of one data set. The test problems for the direct MDR analysis have been simplified by reducing the number of SNPs in each data set to achieve attainable computational time. The displayed time is collected from the processing of multiple independent data sets for each simulation setting. The computational time from the benchmark trial involving 2LOmb, and the combined 2LOmb and MDR approach is the maximum time needed by each method to detect interactions in one data set. In contrast, the computational time from the direct MDR analysis is the minimum time for the completion of interaction detection in one data set. The computational time required by 2LOmb for the null data problem is similar to that for the two-locus interaction problem.

ISSN: 1471-2105