It is accepted that some chromosomal regions, due to a higher number of recombination events, have less consolidated frequency panels in markers underlying these regions in human populations [4]. Allelic frequencies not well defined, or varying between populations, could significantly perturb inferences based on imputed markers in an association study. We investigated a possible relationship between specific chromosomal regions and false discovery events using imputed frequencies. The minus log transformed empiric and imputed P-values and the observed bias between associations statistics were plotted against their relative chromosomal positions (Figure 2) (See methods for further information). The term bias in this analysis refers to the algebraic difference between minus log transformed P-values determined by directly genotyping and imputation. The analysis of Figure 2c suggests the existence of genomic regions more prone to show major biases towards the alternative hypothesis of association when imputation methods were applied. Specifically, polymorphisms located at chromosomes 1, 3 and 15 showed the largest bias in favor of imputed measures. The most prominent biases are concentrated in imputed markers considered strongly associated (p < 10-10) to diabetes
