Should the markers on X chromosome be used for genomic prediction?

Abstract

This study investigated theaccuracy of imputation from LD (7K) to 54K panel and compared accuracy ofgenomic prediction with or without the X chromosome information, based on data ofNordic Holstein bulls. Beagle and Findhap were used for imputation. Averagedover two imputation datasets, the allele correct rates of imputation usingFindhap were 98.2% for autosomal markers, 89.7% for markers on the pseudoautosomal region of the X chromosome, and 96.4% for X-specific markers. Theallele correct rates were 98.9%, 91.2% and 96.8%, respectively, when usingBeagle. Genomic predictions were carried out for 15 traits based on 54K markerdata, imputed 54K for test animals, and imputed 54K for half of referenceanimals. GBLUP models with or without residual polygenic effect were used forgenomic prediction. For all three data sets, genomic prediction using allmarkers gave slightly higher reliability than prediction excluding the X chromosome.Averaged over 15 traits, the gains in reliability from the X chromosome rangedfrom 0.3% to 0.5% points among the three data sets and models.  Using a model with a G-matrix accounting for sex-linkedrelationship appropriately or a model which divided genomic breeding value intoan autosomal component and an X chromosomal component did not led to betterprediction based on the present data where all animals were bulls. A modelincluding polygenic effect did not recover the loss of prediction accuracy dueto exclusion of the X chromosome. It is recommended using markers on the X chromosomefor routine genomic evaluation.