Estimating and adjusting for ancestry admixture in statistical methods for relatedness inference, heritability estimation, and association testing

Abstract

It is well known that genetic association studies are not robust to population stratification. Two widely used approaches for the detection and correction of population structure are principal component analysis and model-based estimation of ancestry. These methods have been shown to give reliable inference on population structure in unrelated samples. We evaluated these two approaches in Mexican American pedigrees provided by the Genetic Analysis Workshop 18. We also estimated identity-by-descent sharing probabilities and kinship coefficients, with adjustment for ancestry admixture, to confirm documented pedigree relationships as well as to identify cryptic relatedness in the sample. We also estimated the heritability of the first simulated replicate of diastolic blood pressure (DBP). Finally, we performed an association analysis with simulated DBP, comparing the performance of an association method that corrects for population structure but does not account for relatedness to a method that adjusts for both population and pedigree structure. Analyses with simulated DBP were performed with knowledge of the underlying trait model.

Figure 1

Individual ancestry analysis. Bar plots of individual-ancestry estimates from a supervised and an unsupervised structure analysis, respectively, with the ADMIXTURE software program for 955 genotyped Genetic Analysis Workshop 18 (GAW18) individuals. A) GAW18-supervised individual ancestry analysis. B) GAW18-unsupervised individual ancestry analysis. Each individual is represented by a vertical bar in A and B. The 4 ancestral populations in B that are inferred by ADMIXTURE in the unsupervised analysis are represented by the colors blue, red, green, and black. The order of the individuals is the same for A and B.

Figure 2

Principal components analysis. The top two principal components from R-PCA (A) and sPCA (B) are plotted against each other. The color of each point in the figures corresponds to an individual's ADMIXTURE-estimated ancestry.

Figure 3

Quantile-quantile (Q-Q) plot for EMMAX and PLINK association analysis with simulated diastolic blood pressure (DBP). Q-Q plots of p-values from EMMAX and PLINK with the top 10 PCs included as covariates for the first simulated replicate of DBP, plotted on the -log₁₀ scale. Red and blue circles in the figure correspond to the association results for PLINK and EMMAX, respectively.

Figure 4

Association results for EMMAX with simulated diastolic blood pressure (DBP). The Manhattan plot of p-values from EMMAX for the first simulated replicate of DBP is plotted on the -log₁₀ scale. The x- and y-axes show chromosome number and -log₁₀ (p-value), respectively.