Scholarlink

<p>Genomic selection (GS) is transforming plant breeding by leveraging statistical machine learning models trained on reference populations that combine phenotypic and genotypic information to predict the performance of new lines based solely on their genotypes. To maximize the efficiency of available resources, various sparse testing strategies have been proposed. In this study, we hypothesize that applying Dynamic Mode Decomposition (DMD) within a sparse testing framework will enhance the prediction of genotype performance in a leave-one-environment-out (LOEO) cross-validation scheme compared with the conventional genomic best linear unbiased prediction (GBLUP) model. To test this hypothesis, we evaluate the effectiveness of the DMD approach against the GBLUP model within a Bayesian framework.Our results indicate that the DMD method consistently outperforms GBLUP model, showing improvements of 76.00% in normalized root mean square error (NRMSE), 27.23% in the percentage of matching of top 10% of lines (PM_10), and 13.37% in the top 20% of lines (PM_20). The percentage of matching of top lines refers to the proportion of genotypes that are simultaneously identified as top performers by both the observed (true) values and the predicted values from a statistical machine learning model. However, in terms of Person´s correlation a decline of 10.88% was observed. The DMD method offers a powerful alternative for sparse testing in genomic prediction by capturing latent patterns directly from phenotypic data without requiring genomic information. This makes it particularly valuable when genomic resources are limited or when evaluating genotypes across novel environments. Although the results are not definitive, they provide supporting evidence that DMD method can enhance prediction accuracy in a sparse testing context compared to conventional GBLUP model across environments.</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

An alternative approach for predicting genotype performance in novel environments without genomic data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月21日

查看全文
  • Osval Antonio Montesinos-López,
  • Kismiantini,
  • Abelardo Montesinos-López

摘要

Genomic selection (GS) is transforming plant breeding by leveraging statistical machine learning models trained on reference populations that combine phenotypic and genotypic information to predict the performance of new lines based solely on their genotypes. To maximize the efficiency of available resources, various sparse testing strategies have been proposed. In this study, we hypothesize that applying Dynamic Mode Decomposition (DMD) within a sparse testing framework will enhance the prediction of genotype performance in a leave-one-environment-out (LOEO) cross-validation scheme compared with the conventional genomic best linear unbiased prediction (GBLUP) model. To test this hypothesis, we evaluate the effectiveness of the DMD approach against the GBLUP model within a Bayesian framework.Our results indicate that the DMD method consistently outperforms GBLUP model, showing improvements of 76.00% in normalized root mean square error (NRMSE), 27.23% in the percentage of matching of top 10% of lines (PM_10), and 13.37% in the top 20% of lines (PM_20). The percentage of matching of top lines refers to the proportion of genotypes that are simultaneously identified as top performers by both the observed (true) values and the predicted values from a statistical machine learning model. However, in terms of Person´s correlation a decline of 10.88% was observed. The DMD method offers a powerful alternative for sparse testing in genomic prediction by capturing latent patterns directly from phenotypic data without requiring genomic information. This makes it particularly valuable when genomic resources are limited or when evaluating genotypes across novel environments. Although the results are not definitive, they provide supporting evidence that DMD method can enhance prediction accuracy in a sparse testing context compared to conventional GBLUP model across environments.