For each individual we defined the HLA genotype at each locus as the pair of alleles with maximum posterior probability as reported by HLA*IMP:02. We performed association analysis (see, for example, ref. 31) for HLA alleles and each disease using logistic regression. The risk model (additive, dominant, recessive or general), as described previously31, was used to enable comparison of effect size estimates. For validation and further details, see Supplementary Information section S5. We repeated the analysis, setting genotypes with a maximum posterior probability of <0.7 to missing. No significant differences were observed compared to the full analysis (data not shown). As a negative control, we ran association analyses in the HLA region with imputed HLA alleles for type 2 diabetes (2,849 cases) and myocardial infarction (9,725 cases) in a total of 409,724 individuals and we found no significant associations (all P > 2.40 × 10−4, the Bonferroni corrected level of association) with any HLA alleles, which is consistent with the lack of associations in the HLA region in recent analyses of each phenotype44,45
