Functional annotation signatures of disease susceptibility loci improve SNP association analysis

BMC Genomics

Table 4 Functional signatures improve inference for association status in a GWAS of ovarian cancer

Variant	Locus	MAF	LO_F	log(BF_A)	Rank_A	Rank_A+F
rs2072590	2q31	0.34	1.46	8.63	65	59
rs2665390	3q25	0.09	0.77	8.08	77	73
rs10069690	5p15	0.23	0.91	-1.38	1,549,122	651,710
rs11782652	8q21	0.08	0.22	2.98	5,272	6,843
rs7814937	8q24	0.12	1.54	14.61	21	16
rs3814113	9p22	0.30	-0.09	14.01	38	38
rs7084454	10p12	0.31	1.44	1.19	45,616	12,221
rs757210	17q12	0.37	1.74	2.31	11,630	2,411
rs2077606	17q21	0.18	0.70	-0.25	339,456	200,494
rs9303542	17q21	0.27	0.05	3.70	2,276	3,532
rs8170	19p13	0.19	0.82	2.72	7,133	4,179
Mean	True +	0.23	0.87	5.15	178,246	80,143
Median	True +	0.23	0.82	2.98	5,272	3,532
Mean	True −	0.35	0.11	0.37	438,664	517,810
Median	True −	0.36	0.06	0.14	181,116	244,393

Ranks of known associated variants (labeled ‘true +’) tend to improve (i.e. are closer to one) when association and functional data are incorporated in the analysis (Rank_A+F) relative to when only the association data are used (Rank_A) and, hence, are more likely to be studied further. Conversely, ranks of (very likely) unassociated variants (labeled ‘true −’) tend to fall with inclusion of the functional data. The functional data for a given variant is summarized by its ‘functional signature’, defined as the prior log–odds of its association given the functional data (LO_F). Aggregate (mean and median) values are provided for the true + set and the true − set. Ranks are out of approximately 2.5M variants.

ISSN: 1471-2164