Statistical Inference of Signatures of Natural Selection in Human Complex Trait Genetic Architecture

Abstract

Genome-wide association studies have used genotype data from large study cohorts to detect large numbers of genetic variants that are significantly associated with complex traits and diseases. However, these detected variants often make up only a small and biased subset of all trait-affecting variants, and their properties are therefore not representative. Here I present novel statistical methods for inferring properties of the full set of genome-wide trait-affecting variants, show results from applying these methods to a number of different complex traits and diseases, and connect them to evolutionary models. Specifically, I present a method for estimating the frequency-dependence of genetic effects. I quantify the total proportion of genetic effects due to variants of different population frequencies and show that rare genetic variants have larger trait effects on average in all analyzed traits. I explain why this is expected under genome-wide purifying selection and analyze the consistency of the frequency-dependent architecture inference model with evolutionary model predictions. Furthermore, I use the inferred frequency-dependence to assess plausible values of relevant evolutionary parameters. I also develop a statistical method to estimate the genome-wide autocorrelation of causal minor allele effect sizes as a function of genomic distance. I show that neighboring variant effects are systematically anti-correlated using data from a range of complex traits and diseases. A possible interpretation is that linked anti-correlated variant effects cancel each other out and are therefore less affected by natural selection. These variants can hence persist in a population on longer timescales and are more likely to be observed.