Abstract
Recently admixed populations comprise a large portion of the human population worldwide, but are often excluded from Genome-Wide Association Studies (GWAS) due to analytic challenges. Our group has previously developed a local ancestry informed generalized linear model based method, Tractor, for GWAS in admixed samples, which produces accurate ancestry-specific effect sizes and boosts discovery power to identify ancestry-enriched loci. Tractor has been instrumental for elucidating the genetic architecture of complex traits across admixed cohorts, however it operates under an assumption of unrelated samples. As biobanks and other large-scale data sources continue to grow, increasing numbers of closely or cryptically related admixed samples are included. This brings new statistical challenges in conducting GWAS and motivates the timely development of novel tools that can model admixture in various cohort settings. Here, we propose a novel mixed model method, Tractor-Mix, that allows for well-calibrated association studies in datasets containing admixed samples with high degrees of relatedness. Similar to Tractor, our method conducts genetic association tests by leveraging local ancestry to produce more accurate effect sizes and boost power under heterogeneity while effectively controlling false positives. Extensive simulations show this enhanced method is competitive with other state-of-the-art approaches that do not produce ancestry-specific results. Empirical testing of Tractor-Mix on multiple cohorts, including admixed samples from the UK Biobank and Mexico City Prospective Study, highlight the value of the method, identifying ancestry-specific associations. In summary, Tractor-Mix is a powerful association framework that extends the capabilities of current models and will facilitate the inclusion of admixed samples in large-scale GWAS.