Run GigaPRS

Estimate SNP effect sizes from GWAS summary statistics (GigaPRS Step 1). LDAK runs inside this page: files you add from your computer are never uploaded. You need three things: LD references for the ancestries of your GWAS, one or more sets of summary statistics, and a few options.

New to GigaPRS? Load a worked example (UK Biobank height with the European panel), then press Estimate effect sizes.

1 · LD reference (SNP-SNP correlations)

GigaPRS needs SNP-SNP correlations from individuals of the same ancestry as each GWAS. Add one panel per ancestry you use; each panel covers about 1.1 million HapMap3 SNPs and is downloaded once, then kept in your browser.

European (UK Biobank): Default for European GWAS; 10,000 UK Biobank individuals born in the UK.
European (CEU): Alternative European panel.
Finnish: For Finnish GWAS such as FinnGen; Haplotype Reference Consortium genotypes.
African: For African and African-American GWAS; UK Biobank individuals of African ancestry.
East Asian: For East Asian GWAS such as Biobank Japan; UK Biobank individuals of East Asian ancestry.

Add your own folder… An extracted correlations folder with .cors.bim, .cors.bin, .cors.root, .cors.windows (and .cors.annotations for the BLD model).

2 · Summary statistics

Add the GWAS results you want to use: your own files, or any of the 51 UK Biobank traits hosted here. Then tell GigaPRS what role each one plays and which LD panel matches its ancestry.

Add your own files… Plain text or .gz. LDAK format (Predictor A1 A2 Z n) is used as is; common formats (PLINK, REGENIE, BOLT-LMM, SAIGE, GWAS Catalog: e.g. SNP A1 A2 BETA SE N) are converted in your browser. SNPs named by position (e.g. 1:752566, or separate chromosome and position columns, GRCh37) are matched to rs IDs using the LD panel.

Which role should each GWAS have?

Focal: A GWAS of the trait you want a score for. GigaPRS produces one set of effect sizes per focal ancestry, so mark one GWAS as focal for each ancestry you want scores for.
Primary: A further GWAS of the same trait (e.g. a second biobank, or another ancestry for which you do not need separate effect sizes). Always used.
Secondary: A GWAS of a different but genetically correlated trait (e.g. BMI when the target is obesity). Only used if it increases the effective sample size by at least 5%, so adding extra ones does no harm.
Multi-ancestry focal: A single GWAS meta-analysed across ancestries. GigaPRS then produces one set of effect sizes; any other GWAS should be secondary.

Examples. One European GWAS: mark it Focal with the European panel. European and African GWAS of the same trait: mark both Focal, each with its own panel, to get ancestry-specific effect sizes. Obesity plus UK Biobank BMI: obesity Focal, BMI Secondary.

3 · Options

The defaults are recommended for most analyses.

BLD heritability model: lets each SNP's expected contribution depend on 86 functional annotations, which usually improves accuracy (downloads about 250 MB extra per ancestry). Untick to assume all SNPs contribute equally. Sample-size check: stop if a focal GWAS has average sample size below 10,000 when several GWAS are combined (genetic correlations are unreliable below this). MCMC solver: estimate the final effect sizes by MCMC instead of variational Bayes. Several times slower, but can be more accurate when the LD panel closely matches the GWAS sample (e.g. a UK Biobank GWAS with the UK Biobank panel). Prior shape of the effect-size distribution; BayesR suits most traits

4 · Run

Hosted files are downloaded first (about 2.9 GB per LD panel the first time; later runs reuse them). LDAK then runs for roughly 20 minutes per focal ancestry with the BLD model (tested with 1.15 million SNPs: 21 minutes, 3.2 GB memory), less without it. Keep the tab open; your computer is kept awake. The result is an effect-size file to use with LDAK (Step 2) on your own genotypes.

Add correlations and summary statistics to start.

LDAK log

Save log · report a problem