Stratify or adjust? Dealing with multiple populations when evaluating rare variants

BMC Proceedings

Table 2 Q1 subpopulation results (20 meta-analyses, 10 replicates each)

Subpopulation	N	Gene	Chromosome	Median (p)	Min, max
Europe^a	1,560	FLT1	13	1.0 × 10⁻⁴⁵	1.1 × 10⁻⁵², 4.6 × 10⁻³⁷
		KDR	4	1.9 × 10⁻⁴¹	8.7 × 10⁻⁵⁸, 1.0 × 10⁻³²
Asia^b	3,210
Yoruba^c	1,120	FLT1	13	9.5 × 10⁻¹⁷	9.3 × 10⁻²⁵, 4.6 × 10⁻¹³
Luhya^d	1,080	VEGFA	6	4.2 × 10⁻⁶	1.7 × 10⁻¹⁴, 1.7 × 10⁻²

Linear regression results using the Count coding for rare variants in a gene. The median, minimum (min), and maximum (max) p-value for the top causative genes that passed our threshold of 10⁻⁶ for at least 50% of the replicates are listed, plus the top result from the Luhya sample, which does not quite meet the threshold.
^a There were 78 lower ranked false positives (p < 10⁻⁶). The highest ranked false positive had p < 10⁻⁴⁰.
^b There were no genes that passed the significance threshold of 10⁻⁶. Top gene had median p = 0.13.
^c There were 23 lower ranked false positives (i.e., p < 10⁻⁶). The highest ranked had p > 10⁻¹⁴.
^d No genes passed the significance threshold of 10⁻⁶ in at least 50% of the replicates. The second ranked gene had p > 10⁻³.

ISSN: 1753-6561