The evolution of methodologies for genomic prediction

Genomic prediction of genotyped individuals utilizes estimates of the effects of alleles at many loci to predict performance. Estimates of allele substitution effects are commonly obtained from multiple regression linear models. In those models, allele substitution effects tend to be treated as random effects, which shrinks their estimates compared to treating them as fixed effects. A number of alternative models that vary subtly can be used in these endeavors. Bayesian approaches that utilize Markov chain Monte Carlo methods that repeatedly sample allele substitution effects are widely used to fit these models. Genomic prediction of non-genotyped individuals utilizes predicted marker genotypes of non-genotyped individuals as well as covariance information based on pedigree relationships. Genomic prediction methodologies continue to evolve, representing a synthesis of approaches, and this paper documents that evolution.