Leveraging local identity-by-descent increases the power of case/control GWAS with related individuals

TL;DR

This paper introduces a new statistical method, cQLS, that leverages local IBD information to enhance the power of GWAS in related individuals, demonstrated through simulations and a real Alzheimer's disease study.

Contribution

The paper presents the cQLS statistic, which incorporates local IBD to improve GWAS power in related samples, outperforming existing methods.

Findings

cQLS increases GWAS power by over 50% in simulations

Decreased p-values for top SNPs in Alzheimer's GWAS example

Genotyping affected siblings is more efficient than random sampling

Abstract

Large case/control Genome-Wide Association Studies (GWAS) often include groups of related individuals with known relationships. When testing for associations at a given locus, current methods incorporate only the familial relationships between individuals. Here, we introduce the chromosome-based Quasi Likelihood Score (cQLS) statistic that incorporates local Identity-By-Descent (IBD) to increase the power to detect associations. In studies robust to population stratification, such as those with case/control sibling pairs, simulations show that the study power can be increased by over 50%. In our example, a GWAS examining late-onset Alzheimer's disease, the $p$-values among the most strongly associated SNPs in the APOE gene tend to decrease, with the smallest $p$-value decreasing from $1.23\times10^{-8}$ to $7.70\times 10^{-9}$. Furthermore, as a part of our simulations, we reevaluate our expectations about the use of families in GWAS. We show that, although adding only half as many unique chromosomes, genotyping affected siblings is more efficient than genotyping randomly ascertained cases. We also show that genotyping cases with a family history of disease will be less beneficial when searching for SNPs with smaller effect sizes.