Data for all SNPs released by the Genotyping Centers are posted on dbGaP, but we recommend filtering association test results based on missing call rate, duplicate discordance, Mendelian errors, sex differences in allelic frequency and heterozygosity, and MAF. The thresholds vary among studies according to quality metric distributions and genotyping platform. For both the Illumina Human1M and Affymetrix 6.0 arrays, the fraction of SNPs that were either failed by the Genotyping Center QC or flagged for filtering during QA is about 7% without the MAF > 0.01 filter (Table I), while the corresponding figure for the Illumina HumanHap550 array is 2%. The recommended MAF filter level is based on power to detect associations. However, for comparison among studies, the filters in Table I are all set to MAF>0.01 in subjects from the United States with primarily European ancestry. For this ethnic group and MAF criterion, the percent of SNPs lost from the Affymetrix 6.0 and Illumina Human1M arrays after both quality and MAF criteria are applied is about 20%, while that for the Illumina HumanHap550 array is 6%. The table
