Efficient Variant Set Mixed Model Association Tests for Continuous and Binary Traits in Large-Scale Whole-Genome Sequencing Studies

Han Chen, Jennifer E. Huffman, Jennifer A. Brody, Chaolong Wang, Seunggeun Lee, Zilin Li, Stephanie M. Gogarten, Tamar Sofer, Lawrence F. Bielak, Joshua C. Bis, John Blangero, Russell P. Bowler, Brian E. Cade, Michael H. Cho, Adolfo Correa, Joanne E. Curran, Paul S. de Vries, David C. Glahn, Xiuqing Guo, Andrew D. Johnson, Sharon L. R. Kardia, Charles Kooperberg, Joshua P. Lewis, Xiaoming Liu, Rasika A. Mathias, Braxton D. Mitchell, Jeffrey R. O’Connell, Patricia A. Peyser, Wendy S. Post, Alex P. Reiner, Stephen S. Rich, Jerome I. Rotter, Edwin K. Silverman, Jennifer A. Smith, Ramachandran S. Vasan, James G. Wilson, Lisa R. Yanek, Susan Redline, Nicholas L. Smith, Eric Boerwinkle, Ingrid B. Borecki, L. Adrienne Cupples, Cathy C. Laurie, Alanna C. Morrison, Kenneth Rice, Xihong Lin

TL;DR

This paper introduces SMMAT, a new family of statistical tests designed to efficiently identify genetic associations with traits using whole genome sequencing data, even in complex datasets with related individuals and population structure. SMMAT offers a powerful and computationally feasible approach for analyzing rare genetic variants in large-scale studies, a critical step for understanding the genetic basis of common diseases.

What Problem Does This Paper Address?

Whole genome sequencing (WGS) generates vast amounts of data, including millions of rare genetic variants. Identifying which of these variants contribute to human traits and diseases is a major challenge. Traditional methods often focus on single variants, which can be underpowered for detecting the effects of rare variants. Variant-set tests, which group variants (e.g., by gene or genomic region), offer a more powerful approach. However, existing variant-set tests struggle with the computational demands of large-scale WGS studies, especially when dealing with complex sample structures involving population stratification and relatedness. This paper aims to bridge this gap by developing computationally efficient and statistically robust variant-set tests that can handle these complexities.

Methods

The authors propose SMMAT (variant-Set Mixed Model Association Tests), a framework built upon the generalized linear mixed model (GLMM). This framework allows for the simultaneous modeling of population structure, cryptic relatedness, and shared environmental effects. SMMAT includes four specific tests:

  • SMMAT-B (Burden Test): Assesses if the combined burden of rare variants in a set is associated with the trait.
  • SMMAT-S (SKAT): A Sequence Kernel Association Test variant, which aggregates variant effects in a flexible way.
  • SMMAT-O (SKAT-O): An optimal combination of the burden and SKAT tests, aiming to maximize power.
  • SMMAT-E (Efficient Hybrid Test): A novel, computationally efficient hybrid test that combines burden and SKAT statistics, designed to be more efficient than SMMAT-O.

A key innovation is that all SMMAT tests share a common null model, which only needs to be fitted once per genome-wide analysis, significantly reducing computational overhead. The tests are constructed using shared single-variant scores and their covariance matrices, further enhancing efficiency.

Datasets

The primary dataset used for illustration is from the National Heart, Lung, and Blood Institute's Trans-Omics for Precision Medicine (TOPMed) program. Specifically, the analysis focused on plasma fibrinogen levels in 23,763 individuals. The authors also generated simulated datasets for type I error rate and power simulations, mimicking complex population structures and relatedness.

Key Insights and Ideas

  • Unified Framework for Complex Data: SMMAT provides a single, flexible framework to analyze rare variants in large, complex WGS datasets, which is crucial for real-world genetic studies.
  • Computational Efficiency is Key: The authors emphasize that for large-scale WGS, computational feasibility is as important as statistical power. SMMAT's design, particularly the shared null model, addresses this directly.
  • SMMAT-E: A Promising Hybrid: The development of SMMAT-E offers a computationally efficient way to combine the strengths of burden and SKAT tests, potentially leading to increased power without prohibitive computational costs.
  • GLMMs are Effective: The simulation studies demonstrate that GLMMs are effective in controlling type I error rates even in the presence of population structure and relatedness, validating the chosen modeling approach.

Key Results

  • Type I Error Control: Simulation studies showed that all four SMMAT tests (SMMAT-B, SMMAT-S, SMMAT-O, SMMAT-E) effectively control type I error rates across various scenarios, including continuous and binary traits, with population structure and relatedness.
  • Power: In large sample sizes (10,000 samples), SMMAT-E demonstrated the highest power for both continuous and binary traits across all tested scenarios.
  • TOPMed Fibrinogen Analysis: The application to TOPMed data identified significant associations with fibrinogen levels, with SMMAT-E, SMMAT-S, and SMMAT-O yielding the most significant results in a key genomic region.
  • Computational Speed: SMMAT-B was the fastest, followed closely by SMMAT-S and SMMAT-E. SMMAT-O was significantly slower due to its grid search for optimal parameter combination.

Limitations and Future Work

  • Small Sample Size Accuracy: SMMAT p-values are based on asymptotic distributions, which may not be accurate in very small samples, especially for binary traits or heavily skewed continuous traits. The authors suggest that small-sample inference procedures developed for SKAT could be adapted for SMMAT.
  • Ultra-Rare Variants: For ultra-rare variants (singletons, doubletons) in small test regions, SMMAT-B might be the most appropriate strategy due to its reliance on cumulative minor allele counts.
  • Weighting Strategy: The optimal weighting strategy for rare variant analysis remains an open research question. While SMMAT can accommodate various weighting schemes (e.g., MAF-based, functional annotations), finding the best strategy is an active area.
  • Combinations of p-values: The paper uses Fisher's method to combine p-values for SMMAT-E. The authors note that other combination methods might be explored for potential improvements.
  • Future Extensions: The authors suggest that the SMMAT framework could be extended to accommodate multiple weights and potentially incorporate methods for GRM-free mixed effects models to further reduce computational costs in very large biobank-scale studies.

Summary updated

Abstract

ABSTRACT With advances in Whole Genome Sequencing (WGS) technology, more advanced statistical methods for testing genetic association with rare variants are being developed. Methods in which variants are grouped for analysis are also known as variant-set, gene-based, and aggregate unit tests. The burden test and Sequence Kernel Association Test (SKAT) are two widely used variant-set tests, which were originally developed for samples of unrelated individuals and later have been extended to family data with known pedigree structures. However, computationally-efficient and powerful variant-set tests are needed to make analyses tractable in large-scale WGS studies with complex study samples. In this paper, we propose the variant-Set Mixed Model Association Tests (SMMAT) for continuous and binary traits using the generalized linear mixed model framework. These tests can be applied to large-scale WGS studies involving samples with population structure and relatedness, such as in the National Heart, Lung, and Blood Institute’s Trans-Omics for Precision Medicine (TOPMed) program. SMMAT tests share the same null model for different variant sets, and a virtue of this null model, which includes covariates only, is that it needs to be only fit once for all tests in each genome-wide analysis. Simulation studies show that all the proposed SMMAT tests correctly control type I error rates for both continuous and binary traits in the presence of population structure and relatedness. We also illustrate our tests in a real data example of analysis of plasma fibrinogen levels in the TOPMed program (n = 23,763), using the Analysis Commons, a cloud-based computing platform.
BibTeX
@article{chen2019efficient,
  author = {Han Chen and Jennifer E. Huffman and Jennifer A. Brody and Chaolong Wang and Seunggeun Lee and Zilin Li and Stephanie M. Gogarten and Tamar Sofer and Lawrence F. Bielak and Joshua C. Bis and John Blangero and Russell P. Bowler and Brian E. Cade and Michael H. Cho and Adolfo Correa and Joanne E. Curran and Paul S. de Vries and David C. Glahn and Xiuqing Guo and Andrew D. Johnson and Sharon L. R. Kardia and Charles Kooperberg and Joshua P. Lewis and Xiaoming Liu and Rasika A. Mathias and Braxton D. Mitchell and Jeffrey R. O’Connell and Patricia A. Peyser and Wendy S. Post and Alex P. Reiner and Stephen S. Rich and Jerome I. Rotter and Edwin K. Silverman and Jennifer A. Smith and Ramachandran S. Vasan and James G. Wilson and Lisa R. Yanek and Susan Redline and Nicholas L. Smith and Eric Boerwinkle and Ingrid B. Borecki and L. Adrienne Cupples and Cathy C. Laurie and Alanna C. Morrison and Kenneth Rice and Xihong Lin},
  title = {{Efficient Variant Set Mixed Model Association Tests for Continuous and Binary Traits in Large-Scale Whole-Genome Sequencing Studies}},
  journal = {The American Journal of Human Genetics},
  year = {2019},
  doi = {10.1016/j.ajhg.2018.12.012},
}

Loading related content...

Figure