Imputation of SNVs and Classical HLA Alleles
Any SNVs showing significant association were checked (e.g., by visual inspection of the intensity cluster plots and investigation of consistency of LD with surrounding markers) and those SNVs deemed unreliable were removed. The final data sets of high-quality SNVs were prephased with SHAPEIT31 and subsequently used to perform imputation with IMPUTE2,32 the 1000 Genomes reference panel (integrated variant set, release March 2012).33 In the Irish AD collection (Table S1), case and control subjects were genotyped on different platforms, and therefore only the 131,692 SNVs in common between the platforms were used to inform imputation.
Postimputation SNVs with low imputation quality (info score < 0.4), call rate <95%, deviation from pHWE < 10−8, or MAF < 5% were excluded. A final data set of approximately 5.2 million SNVs in 2,079 AD case subjects, 3,867 control subjects, 4,212 psoriasis case subjects, and 8,032 control subjects were eligible for subsequent analysis (Table S3).
Classical alleles for HLA-A, HLA-B, and HLA-C were imputed for each case-control cohort separately by HLA∗IMP34,35 and best guess genotypes with probability >0.9. Additional classical HLA-DQA1, HLA-DQB1, and HLA-DRB1 alleles were imputed in each case-control cohort with the exception of the Irish samples, in which there were insufficient informative SNPs. Alleles with a frequency >1% were put forward for analysis. For each individual, alleles were coded as having no, one, or two copies of the respective allele via allele probability >0.9. We obtained high-quality data at the four-digit level with call rates of 92%–100% and accuracy of 92%–98%.