The HWE test appears to detect different types of genotyping artifact on the two genotyping platforms. Figure 4 shows the HWE p-value versus minor allele frequency for the Addiction and T2D NHS projects. The pattern of extreme HWE deviations is strikingly different. The Illumina data (Addiction) show a curve of low p-values that corresponds to SNPs for which one homozygous class is missing, or nearly so (as indicated by the theoretical plot in Figure S12 and the color-coding in Figure 4). This feature is not observed in the Affymetrix data (T2D NHS) and is likely due to the Illumina calling algorithm setting a limit on the distance between adjacent clusters (which may cause merging of adjacent clusters). The extreme HWE deviations in the Affymetrix data show a different pattern: SNPs with relatively low minor allele frequencies tend to have more very significant deviations than those with high frequency, and there are many SNPs in which the heterozygous class is deficient. This feature may be due to the Birdseed algorithm calling by 96-sample plate, which may make calling genotypes of SNPs
