The first aim involved dataset harmonization. Experience has taught us that unless this is conducted with expertise and great care, inference is not secure. Harmonization and quality control apply to each step of the GWAS process – ascertainment of subjects, diagnostic procedures, genotyping, removal of subjects and SNPs with unconfident data, and with extensive searches for bias. [pfs1]For the PGC, raw individual-level and de-identified phenotype and genotype data from each study were uploaded to a high performance computing cluster and processed through a robust and comprehensive quality control pipeline conforming to best-practice protocols in order to minimize chances of false positive results (e.g., due to population stratification). As the individual studies used different genotyping platforms, the cleaned data were imputed against a widely used panel of data from European subjects (HapMap3) (Altshuler et al. 2010) so that all studies had a common set of genotypes. In addition, considerable efforts were made to harmonize phenotype data by ensuring that all studies used comparable diagnostic constructs and to database item-level data.
