Pre-Generated LD With UKB and 1KG Panels¶

Use this tutorial when LD matrices have already been generated outside CREDTOOLS. A common setup is UK Biobank EUR LD for European GWAS and 1000 Genomes LD for non-European populations.

This path skips credtools prepare. You provide a loci_list.txt whose prefix points to existing summary-statistics, LD-matrix, and LD-map files.

File Layout¶

For every population and locus, keep the three files behind one shared prefix:

data/
- EUR_UKB_chr9_21900000_22100000.sumstats.gz
- EUR_UKB_chr9_21900000_22100000.ld.npz
- EUR_UKB_chr9_21900000_22100000.ldmap.gz
- AFR_1KG_chr9_21900000_22100000.sumstats.gz
- AFR_1KG_chr9_21900000_22100000.ld.npz
- AFR_1KG_chr9_21900000_22100000.ldmap.gz

The prefix for the first EUR row is:

data/EUR_UKB_chr9_21900000_22100000

Do not include .sumstats.gz, .ld.npz, or .ldmap.gz in the prefix column.

Required Locus List¶

Create a tab-separated loci_list.txt:

locus_id    chr start   end popu    cohort  sample_size prefix
chr9_21900000_22100000  9   21900000    22100000    EUR UKB 400000  data/EUR_UKB_chr9_21900000_22100000
chr9_21900000_22100000  9   21900000    22100000    AFR 1KG 90000   data/AFR_1KG_chr9_21900000_22100000

Use the GWAS sample size in sample_size. Do not use the reference-panel sample size unless the reference panel is also the GWAS cohort.

Check Variant Alignment¶

The LD map and LD matrix must use the same variant order. The summary statistics do not need to include every LD variant, but the overlap should be high enough for stable QC and fine-mapping.

The LD map should include:

SNPID   CHR BP  A1  A2  AF2

AF2 is optional, but it enables the MAF comparison in QC. For UKB or 1KG LD, AF2 should be the allele frequency from that reference panel.

Run QC First¶

credtools qc loci_list.txt qc_ukb_1kg --threads 4

Start with:

qc_ukb_1kg/qc.txt.gz
qc_ukb_1kg/chr9_21900000_22100000/expected_z.txt.gz
qc_ukb_1kg/chr9_21900000_22100000/dentist_s.txt.gz
qc_ukb_1kg/chr9_21900000_22100000/compare_maf.txt.gz

High lambda_s, many kriging RSS outliers, many Dentist-S outliers, or low maf_corr usually means the GWAS and LD reference disagree. Check genome build, allele coding, population match, and variant order before fine-mapping.

Run Fine-Mapping¶

For one combined meta-analysis:

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

For a multi-input method that should see the UKB and 1KG-backed rows separately:

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

When to Use `prepare` Instead¶

Use credtools prepare when you have PLINK genotype references but not pre-generated {prefix}.ld.npz and {prefix}.ldmap.gz files. In that case, prepare a genotype config:

{
  "EUR": "/ref/ukb_eur",
  "AFR": "/ref/1kg_afr"
}

Then run:

credtools prepare work/chunks/loci_list.txt genotype_config.json work/prepared

After that, use work/prepared/loci_list.txt for QC and fine-mapping.