Skip to content

Multi-Ancestry Fine-Mapping

Multi-ancestry analysis is useful because LD patterns differ across populations. That can make credible sets smaller, but it also adds choices. This tutorial keeps the choices simple.

Start With One Loci List

Rows with the same locus_id belong to the same region:

locus_id    chr start   end popu    cohort  sample_size prefix
locus_1 1   50000000    50500000    EUR UKBB    400000  data/EUR_UKBB_locus_1
locus_1 1   50000000    50500000    AFR MVP 90000   data/AFR_MVP_locus_1
locus_1 1   50000000    50500000    EAS BBJ 180000  data/EAS_BBJ_locus_1

Each row has its own summary statistics and LD files.

Option A: Combine Everything First

credtools pipeline loci_list.txt results_meta_all \
  --meta-method meta_all \
  --tool susie

Use this when you want one combined result per locus.

This is a good first multi-ancestry run because the output is simple. The tradeoff is that population-specific effects can be harder to see.

Option B: Combine Within Each Population

credtools pipeline loci_list.txt results_by_population \
  --meta-method meta_by_population \
  --tool susie

Use this when you have multiple cohorts per population and want population-level fine-mapping inputs.

Option C: Keep Inputs Separate for a Multi-Input Tool

credtools pipeline loci_list.txt results_multisusie \
  --meta-method no_meta \
  --tool multisusie

This lets a multi-input tool use the rows together instead of collapsing them first.

Other multi-input tools include:

credtools pipeline loci_list.txt results_susiex \
  --meta-method no_meta \
  --tool susiex

Compare the Runs

For each run, compare:

results_*/{locus_id}/credible_sets_summary.txt.gz
results_*/{locus_id}/causal_variants.txt.gz
results_*/{locus_id}/pips.txt.gz

Useful questions:

  • Did the lead variant change?
  • Did the credible set shrink?
  • Are high-PIP variants shared across populations?
  • Did QC flag one population more often than the others?

Watch Heterogeneity

The pipeline computes heterogeneity before meta-analysis and writes summary files in each locus output directory.

If a locus has strong heterogeneity, do not rush to interpret the combined result. Run meta_by_population or no_meta and compare.

A practical default

Run meta_all + susie first for a quick baseline. Then run no_meta + multisusie for loci you care about or loci where QC suggests population differences.