Recently, single-step approaches were proposed to overcome the genotyping limitation for some animals. Having considered high genomic performance and utilizing both genotyped and nongenotyped animals, single-step genomic prediction models became the prevailing tool in genetic evaluations of livestock. The objective of current study was to investigate the role of genetic relationships between the training and validation populations and different genomic architecture with simulated genomic data imputation on performance of Boosting, single-step genomic best linear unbiased prediction (SS-GBLUP) and single-step BayesA (SS-BayesA) methods. For this purpose, genomic populations were simulated to reflect variations in number of QTL (10, 100 and 1000) for 29 chromosomes. To simulate a real condition, we randomly masked markers with 70% missing rate for each scenario; afterwards, hidden markers were imputed using FImpute software, and imputation accuracy was estimated. To estimate genomic breeding values, Boosting, SS-GBLUP and SS-BayesA methods were applied for original and imputed genotypes during G1 and G3 generations. According to results, GEBV accuracy was influenced by the relationships between the training and validation populations for ungenotyped animals higher than genotyped ones in both original and imputed genotypes. In both original and imputed genotypes, Boosting model showed the lowest accuracy for genotyped animals. SS-GBLUP method showed an obvious advantage over SS-BayesA and Boosting methods with the scenarios of high QTL. Generally, the relationships between training and validation populations contributed to GEBV accuracy in the single-step and Boosting analysis, and the advantages of SS-BayesA model was more apparent when the trait was controlled by fewer QTL.
 

Keywords: Genomic Relationship, Genomic Selection, Genomic Accuracy, Machine Learning, Single- Step Method

Full-Text [PDF 489 kb]   (517 Downloads)