Evaluating multi-ancestry genome-wide association methods: Statistical power, population structure, and practical implications

Julie‐Alexia Dias, Tony Chen, Xing Hua, Xiaoyu Wang, Alex A Rodriguez, Ravi Madduri, Peter Kraft, Haoyu Zhang

TL;DR

This paper settles a long-standing debate in genomics: is it better to "pool" diverse genetic data into one large analysis or "meta-analyze" separate ancestry groups? Through extensive simulations and real-world data from over 500,000 individuals, the authors show that pooled analysis consistently offers superior statistical power without increasing false positives, primarily because it better leverages differences in allele frequencies across populations.

What Problem Does This Paper Address?

Genome-Wide Association Studies (GWAS) have historically been "European-centric," with nearly 95% of participants coming from European backgrounds as of 2025. This creates a massive gap in our understanding of health for the rest of the world. As researchers rush to include more diverse biobanks, they face a methodological fork in the road:

  1. Pooled Analysis: Throw everyone into one "pot," adjusting for ancestry using Principal Components (PCs). This maximizes sample size but risks "population stratification"—false positive signals caused by ancestry differences rather than actual biology.
  2. Meta-analysis: Run separate studies for each ancestry (e.g., East Asian, African, European) and then combine the results. This is safer against false positives but often loses power and struggles to handle "admixed" individuals (people with multi-continental heritage).

The authors sought to determine which method is truly optimal for discovering genetic variants in a multi-ancestry world.

Methods

The researchers didn't just run a few tests; they built a rigorous comparison framework:

  • Theoretical Derivation: They developed a mathematical framework for "Asymptotic Relative Efficiency" to prove why one method should outperform the other based on allele frequency variance.
  • Three Strategies: They compared Pooled Analysis, standard Fixed-effect Meta-analysis, and MR-MEGA (a more sophisticated meta-analysis that accounts for ancestry-specific effects).
  • Modeling: They tested both Fixed-effect models (assuming genetic effects are constant) and Mixed-effect models (using REGENIE to account for complex relatedness).
  • Simulation vs. Reality: They used HAPGEN2 and Admix-kit to simulate millions of genomes with known "causal" variants, then validated everything using real data from two of the world's largest biobanks.

Datasets

  • Simulated Data: 600,000 subjects for continuous traits and a massive 5-million-person dataset for binary traits (to mimic diseases like breast cancer).
  • Admixed Simulations: Specifically modeled EUR/AFR mixtures to see how methods handle complex ancestry.
  • UK Biobank (UKB): ~324,000 individuals (mostly European, but with significant minority cohorts).
  • All of Us Research Program (AoU): ~207,000 individuals (highly diverse, with ~41% non-European participants).
  • Traits: 13 different traits, ranging from height and cholesterol to asthma and type II diabetes.

Key Insights and Ideas

  • The "Allele Frequency" Advantage: The biggest "aha" moment is that pooled analysis doesn't just win because of a larger N. It wins because it captures the variance in allele frequencies across groups. If a variant is rare in Europeans but common in Africans, the pooled analysis uses that "spread" to find the signal more easily than a meta-analysis, which looks at the groups in isolation.
  • The False Positive Myth: Many researchers avoid pooled analysis for fear of population stratification. However, the authors found that for most real-world traits (where ancestry explains <1% of the trait's variance), pooled analysis is perfectly safe and well-controlled.
  • Admixture is an Asset: Pooled analysis naturally handles individuals with mixed ancestry, whereas meta-analysis requires "binning" people into discrete groups, which is increasingly difficult and scientifically inaccurate in a globalized world.

Key Results

  • Superior Power: In the diverse All of Us dataset, pooled analysis identified 63% more independent significant variants than standard meta-analysis.
  • UK Biobank Gains: Even in the less diverse UK Biobank, pooling found 8% more variants.
  • Mixed-Effect Wins: Using mixed-effect models (like REGENIE) consistently boosted discovery by 6% to 18% compared to simpler fixed-effect models.
  • MR-MEGA Performance: While MR-MEGA is better than standard meta-analysis when ancestry groups are large (>20,000 people), it still lagged behind the pooled approach in almost every scenario.

Limitations and Future Work

  • Homogeneity Assumption: The study assumes that a genetic variant has the same effect size across all ancestries. While often true, some variants might act differently depending on the genetic or environmental background (heterogeneity).
  • Data Privacy: Pooled analysis requires "individual-level data" (sharing the raw genotypes). In many international collaborations, researchers can only share "summary statistics," making meta-analysis the only legal or practical option.
  • Future Directions: The authors suggest developing "hybrid" methods that combine the power of pooling with the flexibility of ancestry-aware meta-analysis to handle cases where genetic effects do vary between populations.

Summary updated

Abstract

Abstract The increasing availability of diverse biobanks has enabled multi-ancestry genome-wide association studies (GWAS), enhancing the discovery of genetic variants across traits and diseases. However, the choice of an optimal method remains debated due to challenges in statistical power differences across ancestral groups and approaches to account for population structure. Two primary strategies exist: (1) Pooled analysis, which combines individuals from all genetic backgrounds into a single dataset while adjusting for population stratification using principal components, increasing the sample size and statistical power but requiring careful control of population stratification. (2) Meta-analysis, which performs ancestry-group-specific GWAS and subsequently combines summary statistics, potentially capturing fine-scale population structure, but facing limitations in handling admixed individuals. Using large-scale simulations with varying sample sizes and ancestry compositions, we compare these methods alongside real data analyses of eight continuous and five binary traits from the UK Biobank (N≈324,000) and All of Us Research Program (N≈207,000). Our results demonstrate that pooled analysis generally exhibits better statistical power while effectively adjusting for population stratification. We further present a theoretical framework linking power differences to allele frequency variations across populations. These findings, validated across both biobanks, highlight pooled analysis as a robust and scalable strategy for multi-ancestry GWAS, improving genetic discovery while maintaining rigorous population structure control.
BibTeX
@article{dias2025evaluating,
  author = {Julie‐Alexia Dias and Tony Chen and Xing Hua and Xiaoyu Wang and Alex A Rodriguez and Ravi Madduri and Peter Kraft and Haoyu Zhang},
  title = {{Evaluating multi-ancestry genome-wide association methods: Statistical power, population structure, and practical implications}},
  journal = {The American Journal of Human Genetics},
  year = {2025},
  doi = {10.1016/j.ajhg.2025.08.006},
}

Loading related content...

Figure