Scholarlink

<p>Forest tree species are in the initial stages of domestication due to their long lifespans, late onset of their sexual phase, and late expression of economically important traits. For these reasons, the breeding populations exhibit large genetic diversity and variable patterns in linkage disequilibrium across genomes. These factors complicate the implementation of genomic selection. Current research indicates that genomic selection (GS) in forest trees is feasible but capturing Mendelian sampling is mostly ineffective due to small family sizes. Poor connectivity among genetic field tests also limits the application of genomic selection in breeding programs, particular where selecting for deployment. We investigated the power of data augmentation to improve the accuracy of predicted genomic estimated breeding values (GEBVs). Synthetic individuals were created by artificial “mixing” of parent genotypes using genomic markers. We tested four different scenarios, where synthetic individuals varied in the level of relatedness to the validation population, with and without inclusion of standard errors during phenotype simulation. Our study found that data augmentation can improve the prediction accuracy of GEBVs. The largest improvement in GEBV accuracy was obtained by increasing genetic relatedness through the generation of “pseudo full-sib” individuals in open-pollinated half-sib populations. A significant improvement in GEBV accuracy was also achieved by generating a number of additional individuals with a similar level of relatedness to the original genotyped populations, especially for traits with low heritability. As a final note of caution, we found that improving prediction accuracy depended on accounting for the standard errors associated with the phenotypes used to generate the synthetic data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data augmentation in pedigreed forest tree population improves genomic prediction accuracy

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Jaroslav Klápště,
  • Mari Suontama,
  • Emily Telfer,
  • Natalie J. Graham,
  • Grahame T. Stovold,
  • Russell McKinley,
  • Charlie Low,
  • Heidi S. Dungey

摘要

Forest tree species are in the initial stages of domestication due to their long lifespans, late onset of their sexual phase, and late expression of economically important traits. For these reasons, the breeding populations exhibit large genetic diversity and variable patterns in linkage disequilibrium across genomes. These factors complicate the implementation of genomic selection. Current research indicates that genomic selection (GS) in forest trees is feasible but capturing Mendelian sampling is mostly ineffective due to small family sizes. Poor connectivity among genetic field tests also limits the application of genomic selection in breeding programs, particular where selecting for deployment. We investigated the power of data augmentation to improve the accuracy of predicted genomic estimated breeding values (GEBVs). Synthetic individuals were created by artificial “mixing” of parent genotypes using genomic markers. We tested four different scenarios, where synthetic individuals varied in the level of relatedness to the validation population, with and without inclusion of standard errors during phenotype simulation. Our study found that data augmentation can improve the prediction accuracy of GEBVs. The largest improvement in GEBV accuracy was obtained by increasing genetic relatedness through the generation of “pseudo full-sib” individuals in open-pollinated half-sib populations. A significant improvement in GEBV accuracy was also achieved by generating a number of additional individuals with a similar level of relatedness to the original genotyped populations, especially for traits with low heritability. As a final note of caution, we found that improving prediction accuracy depended on accounting for the standard errors associated with the phenotypes used to generate the synthetic data.