Use of theoretical and estimated identity-by-descent (IBD) allele sharing measures in genome-wide linkage and association studies, with application to large pedigrees

Abstract

Traditionally, identity-by-descent (IBD) sharing among related individuals is estimated on the basis of the assumed pedigree structure, possibly combined with genotyping information for some or all subjects at a series of genetic markers. Recently, there has been interest in using dense SNP genotype data to estimate both average (across the genome) and local (at particular locations) IBD sharing by pairs of individuals. Although originally intended for inference of pedigree relatedness, these genetically estimated IBDs can potentially replace the traditional IBD estimates used in various genetic data analysis methods. I compared IBD estimates from various software packages (PLINK, KING and linear mixed model (LMM) packages including EMMAX, FaST-LMM, GenABEL, GEMMA and MMM) with the theoretical estimates, and examined their utility in application to LMM association analysis of real and simulated qualitative and quantitative phenotypes from a Brazilian family-based study of visceral leishmaniasis (VL) and from the 18th Genetic Analysis Workshop (GAW) data. Generally, the results from the different software packages were highly concordant. When used to model correlations between individuals in LMM analysis, these approaches achieved good control of type 1 error (well beyond that attainable using theoretical IBD estimates), while also achieving superior power to comparable non- LMM methods. Furthermore, although technically misspecified, LMM methods were also successfully applied to simulated longitudinal data. In addition, a new nonparametric linkage analysis method, Regional IBD Analysis (RIA), is proposed, where theoretical IBD estimates are replaced with the average and local genetic IBD estimates. This method was compared with traditional methods for non-parametric linkage analysis (either exact methods using small pedigrees from a study of vesicoureteral reflux disorder (VUR) or simulation-based methods using large pedigrees from the VL study) and was found to perform at least equally well while taking less time.