Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data, Analytics & AI · METAL Meta-Analysis

METAL: A Command-Line Tool for Meta-Analysis of Genome-Wide Association Studies Using Summary Statistics

METAL was developed in 2007 to combine genome-wide association results from independent studies using summary statistics only. This approach is necessary in human genetics, where pooling raw individual-level genotype data across research groups is frequently precluded by consent and data-sharing restrictions. This article describes the tool’s approach, its weighting methods and the quality control on which its results depend.

Pubrica AcademyWritten by a statistical genetics analyst Reviewed bya genomics methodology reviewer Published10 October 2026 Last checked10 October 2026 Reading time7 minutes Sources11 peer-reviewed

METAL is a command-line software tool, first developed in 2007 by Cristen Willer, Yun Li, and Gonçalo Abecasis, for combining the results of genome-wide association studies (GWAS) conducted by independent research groups. Its central design constraint is that it works entirely with summary statistics (effect sizes, standard errors, and p-values already computed within each individual study) rather than requiring access to raw, individual-level genotype and phenotype data, because such data frequently cannot be pooled across institutions or countries for consent, privacy, and data-governance reasons, even when the researchers involved are fully willing to collaborate.

Learning objectives

  • Why GWAS meta-analysis tools such as METAL use summary statistics rather than pooled raw data
  • The two principal weighting methods offered by METAL, and the circumstances in which each is more appropriate
  • How quality control before meta-analysis affects whether a combined result can be trusted

Common misconceptions

  • “Combining genome-wide association results from different studies is as simple as summing their p-values.” METAL’s documentation describes a sample-size-weighted method as its default approach because p-values alone discard information about the direction of effect and study precision; the tool instead converts each study’s result into a z-statistic that reflects the direction of effect and weights each study’s contribution by the square root of its sample size, which preserves more information than treating p-values as directly poolable values.

Use of summary statistics rather than raw data in GWAS meta-analysis

Willer, Li, and Abecasis's 2010 paper formally describing METAL, published in Bioinformatics, frames the tool as a practical alternative to analyzing a single merged dataset of individual-level genotypes from multiple studies, an approach that is “especially useful when such a direct analysis is impractical,” including situations in which individual-level data cannot be shared across sites1. Peter de Bakker and colleagues, writing specifically on the practical aspects of imputation-driven GWAS meta-analysis, detail how this constraint shapes the entire pipeline: because raw genotype data from different consortia and biobanks are governed by different consent terms, ethics approvals, and national data-protection rules, summary-statistic-based meta-analysis is often the only means of combining results across studies, not simply a convenience2. Eleftheria Zeggini and John Ioannidis, reviewing meta-analysis methodology in GWAS specifically, note that this summary-statistic approach also scales more easily to very large numbers of contributing studies than individual-level pooling would, since each site needs to share only a compact results file rather than its full underlying dataset3. This governance constraint is often unfamiliar to researchers from fields without similarly strict human-subjects data rules, and it should therefore be stated explicitly: the summary-statistics approach is not a simplified or lower-quality substitute for “real” individual-level analysis chosen for convenience, but frequently the only legally and ethically permissible means of combining genetic association evidence across institutional and national boundaries. This is also why the field's major discovery consortia are organized around standardized summary-statistics exchange formats rather than around centralized pooling of raw genotype data from every participating site. A tool that reliably automates this exchange and combination step therefore addresses a governance and coordination problem as much as a statistical one.

The two principal weighting methods in METAL and their appropriate use

METAL's documentation describes two principal ways of combining results across studies for a given genetic marker4. The default, sample-size-weighted method converts each study's result into a z-statistic derived from its p-value and the direction of its effect, then computes an overall z-statistic as a weighted sum across studies, with weights proportional to the square root of each study's sample size and scaled so the squared weights sum to 1.04. The alternative, invoked with the STDERR scheme, instead weights each study's effect-size estimate by the inverse of its standard error, which requires that effect sizes be reported on a comparable scale across studies but can make fuller use of the magnitude of effect rather than only its direction and significance4. Bo Han and Eleazar Eskin developed a further refinement, a random-effects model specifically aimed at GWAS meta-analysis, arguing that a model allowing the true effect to vary across studies, rather than assuming a single fixed true effect across all studies, better reflects the genuine heterogeneity that can arise from differences in population ancestry, genotyping platform, or phenotype definition across contributing cohorts5.

Quality control before pooling as a determinant of reliability

Thomas Winkler and a large international group of co-authors published a detailed protocol specifically on quality control for GWAS meta-analyses, arguing that careful checking of each contributing study's summary statistics before pooling (confirming consistent allele coding and strand orientation, checking for unusual patterns suggestive of genotyping or imputation errors, and verifying genomic control statistics) is as important to the reliability of the final result as the choice of weighting method itself6. Bogdan Pasaniuc and Alkes Price, reviewing methods for dissecting complex-trait genetics using summary association statistics more broadly, make a related point: because the entire meta-analysis pipeline depends on each contributing study's summary statistics being correctly computed and formatted in the first place, errors introduced upstream, in a single study's own analysis, propagate directly into the combined result in a way that is often difficult to detect after the fact without the kind of systematic quality-control checks Winkler and colleagues describe7. Orestis Panagiotou and John Ioannidis's empirical work on replicating borderline genetic associations adds a further caution relevant at the interpretation stage: findings that cross a genome-wide significance threshold only narrowly in a meta-analysis replicate inconsistently in follow-up studies more often than findings that clear the threshold with a substantial margin, a pattern that should be considered when judging the confidence warranted by a specific pooled result8.

Genomic control and correction for inflation

Before any individual study's summary statistics are combined in METAL, one prior check is particularly important: whether that study's own test statistics show evidence of systematic inflation, a uniform upward bias across the entire genome rather than a signal concentrated at genuinely associated markers, which would otherwise be mistaken for widespread true association. Bernie Devlin and Kathryn Roeder developed the genomic control method specifically to detect and correct for this kind of inflation, which commonly arises from undetected population stratification, cryptic relatedness among study participants, or technical artifacts in genotyping, by using the distribution of test statistics across the whole genome, most of which are expected to show no true association, to estimate and remove a uniform inflation factor before interpreting any specific result11. METAL's own documentation includes a dedicated genomic control option for exactly this reason, allowing each contributing study's statistics to be corrected for its own estimated inflation factor before being weighted and combined with the others, which prevents a single poorly controlled study from inflating the apparent significance of the pooled meta-analysis result4. Omitting this check is a specific, documented route by which a GWAS meta-analysis can produce a large number of apparently significant but ultimately spurious associations, which is why Winkler and colleagues' quality-control protocol treats genomic control inspection as a standard item to confirm for every contributing study, not an optional refinement for studies that appear unusual on inspection.

Contributions of large-scale pooling

The principal practical benefit of combining summary statistics across many independent cohorts is a large increase in statistical power to detect genuine but small genetic effects that no single study would be large enough to detect reliably on its own. Cristen Willer and a large international consortium used this kind of summary-statistics meta-analysis approach to identify and refine dozens of genetic loci associated with blood lipid levels, a study whose scale, drawing on data from a large number of contributing cohorts, would not have been feasible through pooling raw individual-level data across every participating site9. The University of Michigan's own documentation for METAL, maintained alongside the tool itself, continues to serve as the primary technical reference for how each weighting scheme and quality-control option in the software is implemented, which is the level of detail a research team requires when configuring a specific meta-analysis run, beyond an understanding of the underlying statistical concept10. The broader lesson from this body of work is that the statistical power gained from summary-statistics meta-analysis is only as reliable as the quality control applied before pooling and the care taken in choosing a weighting method appropriate to the data at hand; a large sample size assembled across many cohorts cannot compensate for allele-coding mismatches, uncorrected inflation, or a mismatched genetic model carried into the pooled result unexamined.

Key points

Key points

METAL’s summary-statistics approach to GWAS meta-analysis exists primarily because pooling raw individual-level genetic data across institutions is frequently precluded by consent and governance requirements; the reliability of its output depends as much on upstream quality control of each contributing study’s statistics as on the choice between its sample-size-weighted and standard-error-weighted methods.

Practical application

Practical application

  1. Allele coding and strand orientation should be confirmed as consistent across all contributing studies before pooling. This is one of the specific quality-control checks shown to materially affect the reliability of a combined GWAS meta-analysis result.
  2. The weighting method should be selected according to what the input data support. The sample-size-weighted default should be used when effect sizes are not on a directly comparable scale across studies, and the standard-error-weighted method when they are.
  3. A borderline genome-wide significant association should be interpreted with appropriate caution. Findings that only narrowly cross the significance threshold in a meta-analysis have been shown to replicate less consistently than findings with a substantial margin, which should inform the strength with which the result is stated.

Frequently asked questions

Consent terms, ethics approvals, and national or institutional data-protection rules frequently prevent individual-level genetic data from being shared or merged across research groups, even when every group involved wishes to collaborate, which is the specific practical problem that summary-statistics-based tools such as METAL were built to address.

The default sample-size-weighted method uses each study's p-value and effect direction, weighted by the square root of sample size; the standard-error-weighted alternative uses each study's actual effect-size estimate weighted by the inverse of its standard error, which requires effect sizes to be on a comparable scale across studies but can use more of the available information when that condition is met.

Not reliably if it only narrowly crosses the threshold; empirical work tracking replication specifically found that borderline genome-wide-significant associations replicate less consistently in follow-up studies than associations that clear the threshold by a wide margin.

The tool and its documentation continue to be maintained, and later methodological work, including random-effects extensions and detailed quality-control protocols, has built directly on it rather than replacing its basic summary-statistics approach, which remains the standard framework for this kind of meta-analysis.

References

  1. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of genomewide association scans. Bioinformatics. 2010;26(17):2190-2191. read 10 October 2026. View on Bioinformatics
  2. de Bakker PIW, Ferreira MAR, Jia X, Neale BM, Raychaudhuri S, Voight BF. Practical aspects of imputation-driven meta-analysis of genome-wide association studies. Human Molecular Genetics. 2008;17(R2):R122-R128. read 10 October 2026. View on Human Molecular Genetics
  3. Zeggini E, Ioannidis JPA. Meta-analysis in genome-wide association studies. Pharmacogenomics. 2009;10(2):191-201. read 10 October 2026. View on Pharmacogenomics
  4. University of Michigan Center for Statistical Genetics. METAL Documentation. read 10 October 2026. View METAL Documentation
  5. Han B, Eskin E. Random-effects model aimed at discovering associations in meta-analysis of genome-wide association studies. American Journal of Human Genetics. 2011;88(5):586-598. read 10 October 2026. View on American Journal of Human Genetics
  6. Winkler TW, Day FR, Croteau-Chonka DC, et al. Quality control and conduct of genome-wide association meta-analyses. Nature Protocols. 2014;9(5):1192-1212. read 10 October 2026. View on Nature Protocols
  7. Pasaniuc B, Price AL. Dissecting the genetics of complex traits using summary association statistics. Nature Reviews Genetics. 2017;18(2):117-127. read 10 October 2026. View on Nature Reviews Genetics
  8. Panagiotou OA, Ioannidis JPA. What should the genome-wide significance threshold be? Empirical replication of borderline genetic associations. International Journal of Epidemiology. 2012;41(1):273-286. read 10 October 2026. View on International Journal of Epidemiology
  9. Willer CJ, Schmidt EM, Sengupta S, et al. Discovery and refinement of loci associated with lipid levels. Nature Genetics. 2013;45(11):1274-1283. read 10 October 2026. View on Nature Genetics
  10. Evangelou E, Ioannidis JPA. Meta-analysis methods for genome-wide association studies and beyond. Nature Reviews Genetics. 2013;14(6):379-389. read 10 October 2026. View on Nature Reviews Genetics
  11. Devlin B, Roeder K. Genomic control for association studies. Biometrics. 1999;55(4):997-1004. read 10 October 2026. View on Biometrics

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 10). METAL: A Command-Line Tool for Meta-Analysis of Genome-Wide Association Studies Using Summary Statistics. Pubrica. https://pubrica.com/academy/concepts-definitions/metal-meta-analysis/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

GWAS Meta-Analysis Pre-Pooling Checklist

This checklist should be completed before summary statistics from multiple GWAS cohorts are combined.

[ ] Confirmed consistent allele coding and strand orientation across all contributing studies
[ ] Checked each study's summary statistics for signs of genotyping or imputation error
[ ] Verified genomic control statistics for each contributing study
[ ] Chosen a weighting method (sample-size vs. standard-error) matched to what the input data supports
[ ] Considered a random-effects model if contributing cohorts differ meaningfully in ancestry or phenotype definition
[ ] Flagged any borderline genome-wide-significant results for cautious interpretation pending replication

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica’s statistical genetics team conducts the full quality-control sequence before any GWAS meta-analysis in METAL, checking allele coding, genomic control and the choice of weighting method so that the pooled result is defensible under review.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Quality control before pooling in GWAS meta-analysis

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.