MVP | Meta-analysis

Overview

Meta-analysis starts from existing GWAS summary statistics. Use this workflow when each cohort or population has already been analyzed separately.

Requires: at least two summary statistics files. See GWAS Summary Statistics File.

mvp --meta \
    --sumstat cohort1.sumstat,cohort2.sumstat,cohort3.sumstat \
    --meta-method 2 \
    --maf 0.01 \
    --out result/meta_demo

Output: combined association results under the selected --out prefix.
Use comma-separated paths in --sumstat; do not insert spaces around commas.
Use --maf to remove low-frequency markers before combining evidence.

Summary Statistics Input

The file format is defined once in GWAS Summary Statistics File. In meta-analysis commands, pass multiple summary files as a comma-separated list.

--sumstat cohort1.sumstat,cohort2.sumstat,cohort3.sumstat

Pre-check	Reason	Recommended action
Genome build	Position mismatches create false marker alignment.	Convert all cohorts to the same build before meta-analysis.
Allele orientation	Effect directions depend on `a1` and `a2`.	Harmonize effect alleles and frequencies across cohorts.
Sample size	Fixed and random effects depend on reliable standard errors and sample sizes.	Use cohort-specific numeric `n` values.
P-value range	Invalid p-values break p-value aggregation methods.	Ensure `0 < p <= 1` whenever possible.

Pre-check. Harmonize marker IDs, genome build, allele orientation and sample-size columns before combining cohorts.

Meta Method Selection

Selector	Method	Use when
`--meta-method 1`	UCALM	Summary-level layered integration for heterogeneous populations.
`--meta-method 2`	Fixed effect	Cohorts are similar and effect directions are expected to match.
`--meta-method 3`	Random effect	Effect sizes may vary across cohorts.
`--meta-method 4`	MR-MEGA	Allele-frequency structure should be modeled across populations.
`--meta-method 5`	ACAT	Signals may appear in only a subset of cohorts.

UCALM shortcut. Use --ucalm for the full individual-level UCALM workflow. For existing summary statistics, use --meta --meta-method 1.

Meta Method Recipes

Switch methods by changing only --meta-method and the output prefix.

# Fixed effect
mvp --meta --sumstat cohort1.sumstat,cohort2.sumstat --meta-method 2 --out result/meta_fe

# Random effect
mvp --meta --sumstat cohort1.sumstat,cohort2.sumstat --meta-method 3 --out result/meta_re

# MR-MEGA
mvp --meta --sumstat cohort1.sumstat,cohort2.sumstat --meta-method 4 --out result/meta_mrmega

# ACAT
mvp --meta --sumstat cohort1.sumstat,cohort2.sumstat --meta-method 5 --out result/meta_acat

Subgroup GWAS to Meta-analysis

When --pop-class is supplied with GLM or MLM, MVP can run subgroup analyses and then integrate subgroup results through the meta-analysis workflow when the meta module is enabled.

mvp --glm \
    --bfile demo \
    --pheno demo.phe \
    --pop-class demo.pop \
    --out result/subgroup_gwas

mvp --meta \
    --sumstat result/subgroup1.sumstat,result/subgroup2.sumstat \
    --meta-method 1 \
    --out result/subgroup_meta

Use this pattern when known population labels are available. For input-file details, see Population Classification File.

Population labels should match the study design, such as cohort, breed, ancestry group or experimental batch.
Do not use --cluster and --pop-class in the same command.
If subgroup sample sizes are very small, inspect subgroup association results before interpreting combined meta-analysis output.

Meta-analysis Output

Meta-analysis writes association output with marker columns and method-specific p-value or effect columns under the selected output prefix.

Output component	Description	How to inspect it
Marker columns	`SNP`, `CHR`, `POS`, `A1`, `A2`.	Confirm marker alignment and allele orientation.
Method columns	Combined statistics generated by the selected meta method.	Use the p-value column corresponding to the selected method.
Subgroup columns	When `--pop-class` is used, subgroup-specific statistics may be included.	Compare subgroup effects with combined results.

Result review. Large blocks of missing values usually indicate marker mismatch, allele mismatch or aggressive filtering in one or more cohorts.