Differences in region-specific recombination rates may account for the few remaining significant SNPs, as variants in areas of especially high recombination rate may be more challenging to impute accurately regardless of QC. To investigate this, we first examined the QCed unfiltered data and found that when the data were dichotomised into those markers with lower (<1 cm/Mb) and higher (≥1 cM/Mb) recombination rates, there were more significant SNPs present in the higher recombination rate group compared with the lower recombination group (P=1.85 × 10−27, average recombination rates of 12.8 and 3.04, respectively). When we examined the QCed data post-imputation QC, this difference disappeared (P=0.526). This clearly indicates that application of the post-imputation QC filter successfully identifies the majority of significant SNPs with high recombination rates. Therefore, to include recombination rate as an extra filter would not be prudent, for example, using the QCed post-imputation QC filtered data and applying a further filter using a recombination rate threshold of >1 cM/Mb would eliminate 2075 SNPs, only 24 of which are significantly different.
