Estratégias para otimizar a população de estimação para predição genômica em Eucalyptus spp

Abstract

Large-scale genotyping has been of great interest in the search for greater genetic gains by genomic selection use. The predictive capacity is the parameter that allows to verify the success of genomic selection in breeding programs. The size and relationship among individuals of the study population, as well as the genetic structure of the training and validation populations, affect the predictions of genomic selection. Based on the above, this work was carried out with the objective of optimizing the training population size and verify the impact of the population structure among the training and validation populations to obtain the genomic predictions. We evaluated an Eucalyptus breeding population, consisting of 860 individuals, phenotypically evaluated for 13 traits at 24 and 36 months old and genotyped by EUChip60k. The predictive capacity of genomic selection was obtained through the correlation between the parametric values and the values estimated by RR-BLUP. The effect of training population size was verified by partitioning into subgroups ranging from 50 to 800 individuals. The training populations were grouped by Principal Component Analysis (PCA) and Bayesian approach (STRUCTURE) to verify the effect of population structure on genomic predictions. The rapid decay of the linkage disequilibrium with the correction for the population structure indicated a strong effect of the structure in the study population. The increase in predictions of the genomic model was small and not significant in subgroups with more than 300 individuals. Predictive abilities in removing relationships between individuals were drastically reduced, leading to negative estimates. An increase in the predictions mean was verified for all evaluated characteristics for the grouping of families by analysis of main components and Bayesian grouping in 3% and 9%, respectively. A greater number of individuals in the training population leads to greater predictive capacity. Population structure has an important role to optimize training populations, providing greater efficiency of prediction models when there are related individuals in the estimation and validation populations.