The Accuracy and Bias of Single-Step Genomic Prediction for Populations Under Selection

摘要: In single-step analyses, missing genotypes are explicitly or implicitly imputed, and this requires centering the observed using means of unselected founders. If only available for selected individuals, on founder mean is not straightforward. Here, computer simulation used to study an alternative analysis that does require but fits μ g individuals as a fixed effect. Starting with diplotypes from 721 cattle, five-generation population was simulated sire selection produce 40,000 phenotypes, which 1000 sires had genotypes. The next generation 8000 genotyped validation. Evaluations were undertaken (J) without (N) when marker covariates centered; (JC) (C) all imputed centered. Centering did influence accuracy genomic prediction, fitting did. Accuracies improved panel comprised quantitative trait loci (QTL); models JC J accuracies 99.4%, whereas C N 90.2%. When markers in panel, 4 80.4%. panels included QTL, model accuracy, little impact contained markers. populations undergoing selection, recommended avoid bias reduction prediction due selection.