Choosing a Meta Method

The meta method controls what happens before fine-mapping. It is one of the most important choices in a multi-study run.

Quick Choice

If you want...	Use
one combined result	meta_all
one result per population	meta_by_population
a multi-input tool to see each row	no_meta
to debug cohort-specific issues	no_meta

meta_all

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

This combines all rows for a locus, regardless of population or cohort.

Use it when:

• you expect effects to be broadly shared,
• you want a simple first answer,
• your sample sizes are small and you need power.

Be careful when:

• QC shows strong heterogeneity,
• one population has a different lead signal,
• allele frequencies differ sharply and one cohort looks like an outlier.

meta_by_population

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

This combines cohorts within each population, then keeps populations separate.

Use it when:

• you have several cohorts per population,
• you want ancestry-level interpretation,
• you want to compare populations without cohort noise.

no_meta

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

This keeps every row separate.

Use it when:

• you are running multisusie, susiex, or another multi-input tool,
• you want to inspect cohort-specific behavior,
• you do not trust a combined result yet.

A Practical Pattern

For important loci, run two passes:

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

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

Then compare the lead variants and credible sets.

Do not treat one method as always correct

Meta-analysis is a modeling choice. If the result matters, compare methods and read the QC tables before making a claim.