Association studies using empiric or imputed frequencies are very sensitive to low quality markers. Accuracy of the Impute algorithm is significantly reduced in alleles showing small MAFs or low calling posterior probabilities [4]. Initially, we determined the association statistics under different models of inheritance in the complete set of markers (See methods for further discussion). We applied a common combined filtering criteria composed of selecting only markers with calling probabilities higher than 0.95 and MAF (minor allele frequency) higher than 1%. To determine the efficacy of this procedure we used a dispersion plot (Figure 1) comparing the P-values (on a -log10 scale) assuming a multiplicative (log-additive) model of inheritance applied to empiric and imputed frequencies of markers in the complete and filtered dataset. The analysis of Figure 1 suggests that this standard quality control procedure was effective in excluding a group of 66.000 (17%) markers showing a significant difference in the magnitude of the association sign between empiric and imputed measures, these elements could potentially influence subsequent GWAS analysis. The same comparison was carried using the QQ plots graphical representation
