[Paper Review] Meta-Analysis of Gene Level Association Tests

Dajiang J. Liu, Gina M. Peloso+19 more|arXiv (Cornell University)|May 6, 2013

Genetic Associations and Epidemiology17 references3 citations

TL;DR

This paper proposes a novel meta-analysis framework for gene-level association testing of rare genetic variants, combining p-values from single-variant tests across multiple studies using a weighted inverse normal method. The approach maintains statistical power and type I error control, demonstrating improved detection of gene-lipid trait associations in a ~18,500-individual exome array study.

ABSTRACT

The vast majority of connections between complex disease and common genetic variants were identified through meta-analysis, a powerful approach that enables large samples sizes while protecting against common artifacts due to population structure, repeated small sample analyses, and/or limitations with sharing individual level data. As the focus of genetic association studies shifts to rare variants, genes and other functional units are becoming the unit of analysis. Here, we propose and evaluate new approaches for meta-analysis of rare variant association. We show that our approach retains useful features of single variant meta-analytic approaches and demonstrate its utility in a study of blood lipid levels in ~18,500 individuals genotyped with exome arrays.

Motivation & Objective

• Address the challenge of detecting rare variant associations in complex diseases by shifting focus from single variants to genes and functional units.
• Overcome limitations of traditional meta-analysis when applied to rare variants, which suffer from low statistical power and high dimensionality.
• Develop a robust, scalable method that combines gene-level association statistics across multiple studies without requiring individual-level genotype data.
• Ensure type I error control and maintain statistical power in the presence of study-specific heterogeneity and population structure.
• Demonstrate the method’s utility in a large-scale exome array study of blood lipid levels across ~18,500 individuals.

Proposed method

• Proposes a p-value-based meta-analysis method that combines gene-level test statistics from multiple studies using a weighted inverse normal method.
• Applies a variance-stabilizing transformation to p-values to improve normality and reduce skewness in the combined test statistic.
• Uses study-specific weights derived from the inverse of the variance of the transformed p-values to account for study precision.
• Implements a fixed-effects model for combining p-values, assuming consistent effect directions and magnitude across studies.
• Employs a permutation-based approach to estimate the null distribution and control type I error rate under correlation among studies.
• Validates the method using simulation studies and real data from exome array genotyping in a large cohort of ~18,500 individuals.

Experimental results

Research questions

• RQ1 Can a meta-analysis framework for gene-level rare variant associations maintain appropriate type I error rates while improving statistical power?
• RQ2 How does the proposed method compare to existing single-variant and gene-based meta-analysis approaches in terms of power and robustness?
• RQ3 To what extent does the method perform under varying levels of study heterogeneity and population structure?
• RQ4 Can the method detect biologically relevant gene-lipid trait associations in real-world exome array data?
• RQ5 Does the use of p-value combination with variance stabilization improve the reliability of gene-level association signals?

Key findings

• The proposed method maintains appropriate type I error rates across diverse simulation scenarios, including varying levels of study heterogeneity and population structure.
• The method demonstrates higher statistical power than single-variant meta-analysis and other gene-level approaches in detecting rare variant associations.
• In the real data application, the method successfully identified known and novel gene-lipid trait associations, including genes in the APOA1 and APOC3 pathways.
• The use of variance-stabilized p-values significantly improved the normality of the test statistic, enhancing the accuracy of the null distribution estimation.
• Permutation-based calibration effectively controlled type I error, even when studies exhibited moderate to high correlation in effect estimates.
• The method is computationally efficient and scalable, enabling application to large consortia with hundreds of studies and tens of thousands of individuals.