Overview

Pan-ancestry GWAS of UK Biobank

Here, we present a multi-ancestry analysis of 7,221 phenotypes using a generalized mixed model association testing framework, spanning 16,119 genome-wide association studies. We provide standard meta-analysis across all populations and with a leave-one-population-out approach for each trait. We develop a stringent quality control pipeline, identifying variants that are discrepant with gnomAD frequencies, and make recommendations for filtering these and other GWAS results.

Multi-ancestry analysis

Participants have been divided into ancestry groups to account for population stratification in GWAS analyses. Throughout these docs, these ancestry groupings are referred to by 3-letter ancestry codes derived from or closely related to those used in the 1000 Genomes Project and Human Genome Diversity Panel, as follows:

  • EUR = European ancestry
  • CSA = Central/South Asian ancestry
  • AFR = African ancestry
  • EAS = East Asian ancestry
  • MID = Middle Eastern ancestry
  • AMR = Admixed American ancestry

These codes refer only to ancestry groupings used in GWAS, not necessarily other demographic or self-reported data.

Release data

We release the summary statistics in two formats:

  • For one or a few phenotypes, we recommend using the phenotype-specific flat files: see further description here.

  • For analysis the full dataset (all phenotypes, all populations), the summary statistics are available in Hail formats: see further description here.

Approach

Analysis was done using SAIGE implemented in Hail Batch to parallelize across populations, phenotypes, and regions of the genome. More details can be found below:

The sample size for each population and the number of phenotypes run is as follows:

PopulationNum. IndividualsNum. Phenotypes
AFR66362493
AMR9801105
CSA88762771
EAS27091612
EUR4205317200
MID15991372

Each phenotype may have fewer samples run, depending on data missingness, which can be found in the phenotype manifest, or n_cases and n_controls in the Hail MatrixTable.