Two molecular measures of relatedness based on haplotype sharing

BACKGROUND: Measuring the extent of shared ancestry between individuals or organisms is important in many fields, including forensic science, conservation genetics and animal breeding. The traditional approach is to calculate the expected degree of relatedness between individuals in a pedigree. This assumes that the founders of the pedigree are non-inbred and unrelated to each other, which is rarely the case. In contrast, molecular data allow measurement of actual relatedness without knowledge of a pedigree. Methods to do this have been proposed, but generally do not take the lengths of the genomic regions shared between individuals into account. RESULTS: Two measures based on the extent of haplotype sharing between genomes are proposed. The intercept measure B estimates the fraction of shared genome between individuals, and the product measure C is closely related to the numerator relationship matrix A. Both are based on a model for the joint distribution of markers at the haplotype level. The two measures are compared to the pedigree-based measure A and to vanRaden’s G, a frequently used molecular measure, using a set of data comprising 5037 dairy cattle. The comparison criteria include the ability to capture genealogical relatedness and the prediction accuracy obtained when used in genomic prediction. Both B and C explain around 95 % of the variation in A, whereas G explains around 6 %. G captures genealogical relatedness poorly, particularly for distantly related individuals (second cousins or farther). Both B and C tend to be larger than A but this can be ascribed to the assumption of non-inbred unrelated founders. Using C in linear mixed models results in slightly higher prediction accuracy than G, and using B results in slightly lower prediction accuracy. CONCLUSIONS: The two proposed measures of relatedness capture genealogical relatedness well, outperforming vanRaden’s G in this respect. When used in genomic prediction models, the product measure leads to slightly improved prediction accuracy.