The genotype data for the included datasets were generated using different platforms, including genotypes called from whole-genome sequencing (AMP-AD, TargetALS12 and GTEx3), genotyping arrays (NABEC11 and Braineac7) and haplotype reference consortium86-imputed genotypes (PsychENCODE datasets), or were called from RNA-seq directly (ENA dataset; Supplementary Note). A total of 22 different genotyping datasets were available, reflecting 6,658 genotype samples (Supplementary Table 1). We performed QC on each dataset separately, using slightly different approaches per platform (Supplementary Note) and used a PCA-based approach to assign ancestries to each individual sample. Most of the included samples were of EUR ancestry: 5,138 samples had an EUR assignment, 805 samples had an AFR assignment and 573 samples were assigned to the other ancestries (Fig. 2b and Supplementary Table 1). We next assessed links between RNA-seq samples and genotyped individuals, and were able to identify 7,644 links (Supplementary Table 1). For eQTL discovery, we grouped these links based on brain region and ancestry, in which we required at least 30 samples spanning at least two cohorts. We next removed sample mix-ups and duplicate samples (Supplementary Note), resulting
