QC Metrics Dictionary

CREDTOOLS QC is meant to answer one practical question: do the summary statistics and LD reference look consistent enough to trust fine-mapping?

Start with qc.txt.gz, then open the detailed files only for suspicious loci.

qc.txt.gz

Column	Meaning	How to read it
popu	population label	comes from loci_list.txt
cohort	cohort label	comes from loci_list.txt
n_snps	variants used in QC	small values can mean poor sumstats/LD overlap
n_1e-5	variants with p-value below 1e-5	quick signal count
n_5e-8	variants with p-value below 5e-8	genome-wide significant count
maf_corr	correlation between sumstats MAF and LD-reference MAF	low values suggest allele-frequency mismatch
lambda_s	RSS regularization estimate	high values suggest LD or summary-stat mismatch
n_lambda_s_outlier	count from LD-mismatch and marginal rules	variants flagged by the kriging RSS rules
n_dentist_s_outlier	count from Dentist-S	variants inconsistent with the lead-SNP LD pattern
n_c1b_outlier	optional C1b count	appears when C1b/adaptive QC is enabled

Look for patterns, not one number

A single odd metric is a prompt to inspect the locus. A pattern across many loci or one cohort is a stronger sign that the input panel, build, or allele coding needs attention.

expected_z.txt.gz

This file comes from the kriging RSS check.

Column	Meaning
SNPID	variant identifier
z	transformed z-score
condmean	expected z-score given other variants and LD
condvar	conditional variance
z_std_diff	standardized residual between observed and expected z-score
logLR	log likelihood ratio for allele-switch-like behavior
lambda_s	locus-level regularization estimate
cohort	combined popu_cohort label

Default C1 and C2 rules use:

Rule	Default condition
C1 LD mismatch	logLR > 2 and abs(z) > 2
C2 marginal	abs(z) < 2, abs(z_std_diff) > 3, and lead-SNP correlation above 0.8
C1b high-z residual	abs(z_std_diff) > 10 and abs(z) > 2

C1b is only counted when enabled with --enable-c1b or through adaptive QC.

dentist_s.txt.gz

Dentist-S asks whether a variant's marginal association is consistent with the lead variant and LD.

Column	Meaning
SNPID	variant identifier
t_dentist_s	Dentist-S test statistic
-log10p_dentist_s	evidence against consistency
r2	LD R2 with the lead variant
cohort	combined popu_cohort label

The default Dentist-S outlier rule is:

-log10p_dentist_s >= 4 and r2 >= 0.6

compare_maf.txt.gz

Column	Meaning
SNPID	variant identifier
MAF_sumstats	minor allele frequency from summary statistics
MAF_ld	minor allele frequency from the LD map AF2 column
cohort	combined popu_cohort label

If the LD map has no AF2 column, CREDTOOLS cannot compute this comparison and the detailed MAF file may be empty.

Outlier Removal Files

When --remove-outlier or --adaptive-qc is used, CREDTOOLS writes a cleaned/ folder.

File	Use
cleaned/outlier_snps.txt.gz	variant-level outlier calls
cleaned/outlier_removal_summary.txt.gz	counts removed per locus and cohort
cleaned/cleaned_loci_info.txt.gz	input file for downstream fine-mapping on cleaned data

--adaptive-qc implies outlier removal. It first removes baseline C1, C2, and C3 outliers. If the cleaned locus still has high lambda_s, it adds C1b outliers from the original locus and recomputes QC.

Heterogeneity Metrics

Meta-analysis writes heterogeneity outputs before combining cohorts.

Metric	File	Meaning
LD fourth moment	ld_4th_moment.txt.gz	local LD structure around each variant
LD decay	ld_decay.txt.gz	average LD by distance bin
SNP missingness	snp_missingness.txt.gz	which cohorts contain each variant
Cochran-Q	cochran_q.txt.gz	effect-size heterogeneity across cohorts
Summary	heterogeneity.txt.gz	per-cohort rollup

Use heterogeneity.txt.gz for the first pass. Open detailed files when one cohort has unusually high missingness, high LD differences, or many heterogeneous SNPs.

Practical Review Flow

1. Sort qc.txt.gz by lambda_s, n_lambda_s_outlier, and maf_corr.
2. Pick the worst locus for each suspicious cohort.
3. Open expected_z.txt.gz, dentist_s.txt.gz, and compare_maf.txt.gz.
4. Plot the locus with credtools plot.
5. Decide whether to remove outliers, change LD reference, or drop the locus.