In testing for possible confounding effects, we observed that the unadjusted most significant SNP per gene p-value is affected by several gene properties, most notably physical gene size and number or density of SNPs per gene, and the genetic properties: number or density of SNPs across a gene that are in linkage equilibrium to each other and number or density of recombination hotspots that span a gene. While gene size and number of SNPs per gene have been recently reported to be correlated with the unadjusted best SNP p-value [23], [25], [33], we have quantitatively demonstrated the magnitude of these and linkage-based effects using randomized GWA study data, confirming their potential confounding effects. We show that large genes tend to receive a more significant score than small genes by chance (Figure S2).
