Estimating Trans-Ancestry Genetic Correlation with Unbalanced Data Resources

Bingxin Zhao, Xiaochen Yang, Hongtu ZhuJournal of the American Statistical Association 2024 · DOI

TL;DR

This paper introduces a new statistical method to estimate how genetic influences on complex traits differ across populations, even when one population has much less genetic data than another. This is crucial for understanding disease mechanisms and improving genetic predictions for underrepresented groups.

What Problem Does This Paper Address?

Genome-wide association studies (GWAS) have revealed genetic links to many complex traits and diseases. However, the genetic architecture of these traits can vary significantly between different ancestral populations. Understanding this "trans-ancestry genetic correlation" is vital for developing effective treatments and interventions that work across diverse groups. A major challenge is the severe imbalance in available GWAS data: most large-scale studies are from European populations, while data from non-European populations are scarce. This makes it difficult to accurately estimate genetic correlations between these groups. Furthermore, differences in linkage disequilibrium (LD) – the non-random association of genetic variants – between populations can bias these estimates. Existing methods often struggle with these issues, particularly when dealing with limited data from one of the populations.

Methods

The authors propose a novel estimator for trans-ancestry genetic correlation that addresses both data imbalance and LD heterogeneity. The core idea is to use genetically-predicted traits from the population with limited data, where the genetic effects are learned from the large-scale GWAS data of the other population. This approach allows for a large GWAS sample in one population (e.g., European) and a much smaller sample in the other (e.g., non-European).

Key methodological contributions include:

  • LD Difference-Aware Estimation: The method explicitly accounts for differences in LD patterns between populations, which is a common source of bias in previous approaches.
  • Correction for Prediction Errors: When using genetically-predicted traits, there are inherent prediction errors. The proposed estimator corrects for these errors, which are particularly pronounced in high-dimensional settings with weak genetic signals.
  • Flexible Sample Size Requirements: The estimator is designed to work effectively even when one population has a very small sample size (e.g., hundreds of individuals), making it suitable for underrepresented groups.
  • Theoretical Framework: The paper provides a rigorous theoretical framework, including detailed mathematical proofs and conditions, to establish the consistency and properties of their proposed estimator.

Datasets

The study utilizes the UK Biobank (UKB) dataset for real data analysis. Specifically, they use genotype data from European (White British) and Asian subjects. For simulations, they generate synthetic genotype data mimicking real-world LD patterns.

Key Insights and Ideas

  • Data Imbalance is a Major Hurdle: The paper strongly emphasizes that the scarcity of genetic data for non-European populations is a critical bottleneck for trans-ancestry genetic research. Their method directly tackles this.
  • LD Heterogeneity is Not Trivial: Differences in LD between populations can significantly distort genetic correlation estimates, even when the underlying genetic effects are similar. Their method's ability to account for this is a key innovation.
  • Genetically-Predicted Traits are Powerful, but Need Correction: Using genetically-predicted traits from one population to study another is a promising avenue, but naive approaches can lead to underestimated genetic correlations due to prediction errors and LD mismatches.
  • The Importance of Correcting for Bias: The corrected estimator consistently outperforms naive methods, especially in scenarios with significant LD differences or limited data.

Key Results

  • Reliable Estimates with Unbalanced Data: Through extensive simulations and real data analyses on 30 complex traits from the UK Biobank, the proposed method demonstrates its capability to provide reliable estimates of trans-ancestry genetic correlations, even with vastly different sample sizes between populations.
  • Improved Accuracy Over Naive Methods: The corrected estimator shows significantly less bias compared to naive methods, particularly in scenarios with substantial LD differences. For instance, in simulations, the naive estimator was often found to be about 5 times smaller than the true genetic correlation, while the corrected estimator was much closer.
  • Insights into Trait Transferability: The real data analysis on UK Biobank data revealed that while genetic architectures are generally similar between European and Asian populations, there are also population-specific differences, impacting the transferability of genetic findings.

Limitations and Future Work

  • Assumption of Polygenic Architecture: While the method is designed to be robust, it relies on a polygenic model where many genetic variants contribute to a trait. For traits with very sparse genetic signals (few causal variants), the variance of the estimator might increase, potentially leading to slight overestimation. The authors suggest incorporating penalty or thresholding approaches for such cases.
  • Reference Panel Choice: The paper explores the use of reference panels for LD information. While they show that matching the reference panel's LD structure to the training population can improve accuracy, the optimal choice of reference panel in all trans-ancestry scenarios remains an area for further investigation.
  • Identifying Specific Loci: The authors suggest future work could focus on identifying specific genetic loci that contribute to genetic differences between populations, especially where genetic effects are heterogeneous.
  • Transfer Learning Applications: The work has implications for transfer learning methods, and further research could explore how to best incorporate this understanding of cross-population genetic architecture into such frameworks.

Summary updated

Abstract

The aim of this paper is to propose a novel method for estimating trans-ancestry genetic correlations in genome-wide association studies (GWAS) using genetically-predicted observations. These correlations describe how genetic architecture of complex traits varies among populations. Our new estimator corrects for biases arising from prediction errors in high-dimensional weak GWAS signals, while addressing the ethnic diversity inherent in GWAS data, such as linkage disequilibrium (LD) differences. A distinguishing feature of our approach is its flexibility regarding sample sizes: it necessitates a large GWAS sample only from one population, while the secondary population may have a much smaller cohort, even in the hundreds. This design directly addresses the existing imbalance in GWAS data resources, where datasets for European populations typically outnumber those of non-European ancestries. Through extensive simulations and real data analysis from the UK Biobank study encompassing 26 complex traits, we validate the reliability of our method. Our results illuminate the broader implications of transferring genetic findings across diverse populations.
BibTeX
@article{zhao2024estimating,
  author = {Bingxin Zhao and Xiaochen Yang and Hongtu Zhu},
  title = {{Estimating Trans-Ancestry Genetic Correlation with Unbalanced Data Resources}},
  journal = {Journal of the American Statistical Association},
  year = {2024},
  doi = {10.1080/01621459.2024.2344703},
}
Figure