A test for linkage disequilibrium in structured populations

Abstract

Quantifying patterns of dependence between distinct loci, known as linkage disequilibrium (LD), is vital for identifying locus-trait associations and studying evolution. LD is well understood when individuals are sampled from a population where Hardy-Weinberg equilibrium (HWE) holds at each locus. However, population structure is nearly ubiquitous in human populations, leading to significant deviations from HWE and inducing correlations between loci that mimic LD. These confounding signals represent an ongoing challenge when characterizing LD for individuals sampled from large human datasets. To address this, we develop a hypothesis test for a quantity we call structural LD, defined as LD after accounting for population structure. Our test accommodates arbitrarily complex structure using individual-specific allele frequencies, which are possibly distinct for every SNP-individual pair. This formalization is highly general and encompasses a wide range of established models, including admixture. We validate the performance of our proposed method on both simulated genotypes and existing human studies, including data characterized by highly complex structure and recent admixture. Our analyses of human studies indicates that most long-range LD in human populations can be accounted for by well-fit models of population structure.