different thresholds of calling probabilities were not efficient predictors any further. When the same plotting routine was used using MAF as a predictor for bias (Figure 3), interesting results could be observed. Most prominent bias was encountered in markers with higher MAF (close to 0,5), suggesting that in this allelic condition the imputation algorithm was jeopardized by difficulties to determine the major and minor allele. This interesting feature was further analyzed to determine if this specific allele condition could be considered a useful predictor to identify these markers. We selected a subset of markers that were directly genotyped and imputed showing extreme MAF conditions (MAF < 0,01 and MAF > 0,49) and compared their transformed association measures with the use of dispersion plots and histograms (Additional file 6, Figure S2). The analysis of these figures show that the vast majority of markers in these allelic conditions have significant agreement for their association statistics but, as determined before, a limited number of markers, especially the ones showing MAF very close to 0,5, have an increased odds of being biased. The barrier is probably imposed by allele misspecification and is a challenging one, since a specific allele could be wrongly imputed leading
