A robust and powerful two‐step testing procedure for local ancestry adjusted allelic association analysis in admixed populations

Abstract

Genetic association studies in admixed populations allow us to gain deeper understanding of the genetic architecture of human diseases and traits. However, population stratification, complicated linkage disequilibrium (LD) patterns, and the complex interplay of allelic and ancestry effects on phenotypic traits pose challenges in such analyses. These issues may lead to detecting spurious associations and/or result in reduced statistical power. Fortunately, if handled appropriately, these same challenges provide unique opportunities for gene mapping. To address these challenges and to take these opportunities, we propose a robust and powerful two-step testing procedure Local Ancestry Adjusted Allelic (LAAA) association. In the first step, LAAA robustly captures associations due to allelic effect, ancestry effect, and interaction effect, allowing detection of effect heterogeneity across ancestral populations. In the second step, LAAA identifies the source of association, namely allelic, ancestry, or the combination. By jointly modeling allele, local ancestry, and ancestry-specific allelic effects, LAAA is highly powerful in capturing the presence of interaction between ancestry and allele effect. We evaluated the validity and statistical power of LAAA through simulations over a broad spectrum of scenarios. We further illustrated its usefulness by application to the Candidate Gene Association Resource (CARe) African American participants for association with hemoglobin levels. We were able to replicate independent groups' previously identified loci that would have been missed in CARe without joint testing. Moreover, the loci, for which LAAA detected potential effect heterogeneity, were replicated among African Americans from the Women's Health Initiative study. LAAA is freely available at https://yunliweb.its.unc.edu/LAAA.

Ancestry-specific association mapping in admixed populations

“…To investigate the cost and benefits of asaMap compared to a standard GLM based test we first applied both methods to simulated data to compare their statistical power and to assess important statistical properties of asaMap, like bias and false positive rates. Using the same simulated data we also compared asaMap to another similar test, the LAAA test recently proposed by Duan et al (2017). Finally, we applied both the standard GLM based methods to real data to compare the range of their potential usage.…”

“…We emphasize that asaMap is an association testing method and not a method for performing admixture mapping; a mapping approach where correlation between phenotype and inferred local ancestry is used to identify candidate regions (Patterson et al 2004). asaMap is more similar to the 7 Skotte Mapping ancestry specific effects methods of Pasaniuc et al (2011), Yorgov et al (2014 and Duan et al (2017),…”

Ancestry‐specific association mapping in admixed populations

“…Like similar methods by Pasaniuc et al (2011), Yorgov et al (2014, and Duan et al (2017), asaMap focuses only on one specific allele, the assumed effect allele, when allowing for ancestry-specific effects, as illustrated in Figure S1A. Thus it ignores the possibility that the other allele, the assumed noneffect allele, could mediate an ancestry-specific effect ( Figure S1B).…”

“…We emphasize that asaMap is an association testing method and not a method for performing admixture mapping; a mapping approach where the correlation between phenotype and inferred local ancestry is used to identify candidate regions (Patterson et al, ). asaMap is more similar to the methods of Pasaniuc et al (), Yorgov, Edwards, and Santorico (), and Duan et al (), where ancestry‐specific effects are estimated and tested based on inferred knowledge about local ancestry. These methods focus on the potentially different effects of one specific allele (which we will refer to as “the assumed effect allele”) in the ancestral populations and they assume that the other allele (which we will refer to as “the assumed noneffect allele”) has the same mean phenotypic value in all ancestral populations after correcting for relevant covariates.…”

Impact of admixture and ancestry on eQTL analysis and GWAS colocalization in GTEx

“…The can be problematic when genes are differentially expressed in ancestral populations of admixed individuals. In contrast, local ancestry (LA), or the number of alleles derived from distinct ancestral populations at a given locus, may be more appropriate for population structure adjustment in admixed populations but typically suffers from much longer compute time and can be prone to errors in estimation at a variant level (5,(8)(9)(10)(11)(12).…”