absolute difference in call rate between sexes is greater than 2.5%, if the absolute difference in heterozygosity between sexes is greater than 7%, or if cluster separation <0.20. Vanderbilt BioVU, Mayo, and Northwestern NUGene samples were genotyped at the Broad Institute, where technical failure was determined by call rate 95%, GenTrain score <0.6 (a statistical measure from Illumina’s clustering algorithm[41]), cluster separation <0.4, or more than one replicate error. It is also advantageous to build in duplicate samples to estimate the reproducibility rate within genotyping batches. By design, both CIDR and Broad include HapMap control samples and duplicate samples across all plates in the study. It is anticipated that for accurate genotyping data, duplicate reproducibility and HapMap concordance of >99% are expected. We removed any SNPs which had one or more discordant calls on duplicate samples. Both HapMap concordance and replicate sample concordance can be checked using the concordance procedure in the PLATO software [42] (Table 1) or by using the --genome --rel-check options in PLINK. Using HapMap trio samples it is also possible to inspect each SNP for Mendelian inconsistencies, which indicate genotyping errors if pedigree information is correct. Mendelian inconsistencies can be assessed using the --mendel option in
