Abstract
Multidimensional scaling is a well-known multivariate technique, that is often used in genetics for studying population substructure. In this paper we show that multidimensional scaling of marker data is of relevance for relatedness research. Relatedness is usually investigated by estimating and plotting identity-by-state and identity-by-descent allele-sharing statistics. We show that outlying individuals in a map obtained by multidimensional scaling of genetic variables do not necessarily stem from a different human population, but can be the consequence of relatedness. We propose a method for classifying pairs of individuals into the standard relationship categories that combines genetic bootstrapping, multidimensional scaling and discriminant analysis. We validate our method with simulation studies. Given the variant filtering procedures, our method classifies relationships up to and including the fourth degree with high accuracy (96-97%), using only identity by state. The usefulness of the method is illustrated with data from the 1,000 genomes and the GCAT projects.
