How Would Different Models of Genetic Variation Affect Genomic Selection?

Abstract

Genomic selection (GS) is a method to predict breeding values in livestock; however the underlying mechanism by which it predicts is not fully clear. Initially it was thought that GS predicted effects of quantitative trait loci (QTL), in linkage disequilibrium (LD) with markers (Meuwissen et al. 2001). Increasingly, there has been a realization that GS predicts 'relationships' between animals (Habier et al., 2010). Literature suggests that a very small part of the additive genetic variance is explained by variation at known QTL (Maher, 2008). Fearnhead et al, (2004) noted that this is often inconsistent with high estimates of heritability and they proposed that a rare variant model might explain this "missing heritability". Given this debate, it is relevant to obtain greater understanding into what GS is actually predicting for several reasons. Firstly, the LD paradigm predicts that GS can estimate breeding values with higher accuracies as long as marker densities are increasing, possibly even allowing prediction of breeding value across breeds (Goddard et al., 2006). Moreover, accurate prediction would persist for several generations into the future. In contrast, if the relationship paradigm is true, then our predictive ability based on genomic data would remain for only one or two generations ahead. Consequently continuous measurement of phenotypes of individuals that are at least somewhat related to selection candidates would be needed.