Statistical models for the prediction of genetic values

Abstract

Agricultural and medical genetics are currently revolutionized by the technological developments in genomic research. The genetic analysis of quantitatively inherited traits and the prediction of the genetic predisposition of individuals based on molecular data are rapidly evolving fields of research. We ask how phenotypic variation for a quantitative trait can be linked to genetic variation at the DNA level. Advances in high-throughput genotyping technologies return data on thousands of loci per individual. We present linear models to identify molecular markers significantly associated with quantitative traits. We discuss the drawbacks arising from a large number of predictor variables and a high degree of collinearity between them. We illustrate how linear mixed models can overcome the limitations through shrinkage and allow the prediction of genetic values inferred from genome-wide marker data. With a small example from maize breeding, we present how the models can be applied to predict the risk of genetically diverse individuals to be damaged by insects and why predictions based on whole-genome marker profiles are likely to be more accurate than those based on pedigree information. The choice of appropriate methods for quantitative genetic analyses based on high-throughput genomic data for medical and agricultural genetics is discussed.