The Relation Between Inflation in Type-I and Type-II Error Rate and Population Divergence in Genome-Wide Association Analysis of Multi-Ethnic Populations

E M Derks,A H Zwinderman,E R Gamazon

doi:10.1007/s10519-017-9837-3

Abstract

Population divergence impacts the degree of population stratification in Genome Wide Association Studies. We aim to: (i) investigate type-I error rate as a function of population divergence (FST) in multi-ethnic (admixed) populations; (ii) evaluate the statistical power and effect size estimates; and (iii) investigate the impact of population stratification on the results of gene-based analyses. Quantitative phenotypes were simulated. Type-I error rate was investigated for Single Nucleotide Polymorphisms (SNPs) with varying levels of FST between the ancestral European and African populations. Type-II error rate was investigated for a SNP characterized by a high value of FST. In all tests, genomic MDS components were included to correct for population stratification. Type-I and type-II error rate was adequately controlled in a population that included two distinct ethnic populations but not in admixed samples. Statistical power was reduced in the admixed samples. Gene-based tests showed no residual inflation in type-I error rate.

Highlights

Genome-Wide Association Study (GWAS) is a widely used approach for the identification of genetic variants that are associated with disease risk (Manolio 2010)
We aimed to test whether correction for population stratification using estimates of global ancestry adequately controls type-I error rate and adequately retains statistical power in ethnically heterogeneous GWAS data
We tested explicitly whether the degree of residual inflation depends on the degree of population divergence (FST), as we hypothesized that population stratification would be more difficult to control for Single Nucleotide Polymorphisms (SNPs) with relatively high values of FST

Summary

Introduction

Genome-Wide Association Study (GWAS) is a widely used approach for the identification of genetic variants that are associated with disease risk (Manolio 2010). As pointed out by Medina-Gomez and colleagues, excluding subjects with different ethnic backgrounds may reduce sample size and thereby lead to a loss of statistical power (Medina-Gomez et al 2015). For these reasons, Medina-Gomez et al aimed to investigate the effect of population admixture on type-I error rate in Genome Wide Association Studies (GWAS) (Medina-Gomez et al 2015). Correction for population stratification was done by including genomic components as covariates in the association model (see, e.g., (Price et al 2006) or alternatively, by using Efficient Mixed-Model Association

Objectives

Methods

Results

Conclusion