A genealogical interpretation of fixed and random effects models of complex traits

Abstract

In the analysis of complex traits, genetic effects are frequently modelled as either fixed or random effects. Such assumptions serve as a foundation of defining heritability and relatedness using genome-wide single nucleotide polymorphism (SNP) markers. In this work, I propose a genealogical framework connecting the two assumptions conditional on the ancestral recombination graph (ARG). It turns out that the reference time point in which the probability is defined determines whether the effect of a variant to behave as either a fixed or random effect. This lays a connection between the PC regression and linear mixed model (LMM) for genetic association study. The framework induces a genetic relatedness matrix (GRM) in which the elements are a function of the time to the recent common ancestor. Subsequently, a novel trait variance decomposition respect to the regions sharing a common genealogy is followed. The variance decomposition then provides a natural means to define heritability and a potential method for gene mapping.