Chi8: a GPU program for detecting significant interacting SNPs with the Chi-square 8-df test

BMC Research Notes

Table 6 Total Chi8 and GBOOST time in hours to run on the seven WTCCC GWAS dataset

Data	T1D	RA	CD	CAD	T2D	HT	BD
Time
Chi8	14.2	10.1	8.4	8.6	8.6	8.7	8.3
GBOOST	1.7	1.6	1.6	1.6	1.6	1.6	1.5
Number of significant pairs
Chi8	1644	11	4	0	0	0	0
GBOOST	28K	11K	9.9K	10K	10K	10K	9.8K
GBOOST*	92	0	0	0	0	0	0
BOOST (published [14])	91	0	1	0	0	0	0
Number of common pairs between Chi8 and GBOOST
	236	2	4	NA	NA	NA	NA

We also show number of significant pairs outputted by each program such that both SNPs are insignificant on their own. Otherwise our program Chi8 outputs thousands of interactions where one SNP is significant on its own. For GBOOST we use the same interaction threshold of 37 as we did for the simulated data. We also constrain the GBOOST output (denoted as GBOOST*) to consider just pairs with a threshold above 175 and at least 100 SNPs apart to account for linkage. We selected this threshold so as to output a similar number of pairs for type 1 diabetes as published in the original BOOST paper [14]. The previously published values from the original BOOST study consider pairs that are at least 1MB apart and individually insignificant [14]

ISSN: 1756-0500