Multi-generation genomic prediction of maize yield using parametric and non-parametric sparse selection indices.

Area of Science:

• Quantitative genetics
• Plant breeding
• Genomic prediction

Background:

• Genomic prediction models often use multi-generation data, leading to heterogeneous training sets.
• Allele frequency and linkage disequilibrium differences can limit prediction accuracy.
• Optimizing training set selection is crucial for accurate genomic predictions.

Purpose of the Study:

• To investigate if combining sparse selection index (SSI) with kernel methods improves prediction accuracy in genomic models trained on multi-generation data.
• To compare the performance of kernel-based SSI against traditional genomic best linear unbiased prediction (GBLUP) and non-parametric kernel methods (KBLUP).

Main Methods:

• Utilized four years of doubled haploid maize data from CIMMYT.
• Applied sparse selection index (SSI) for training set optimization.
• Compared additive genomic relationships (GSSI) and Gaussian kernel methods (KBLUP) for prediction.

Main Results:

• Kernel-based methods (KBLUP) outperformed GBLUP for predicting grain yield.
• Genomic SSI (GSSI) using additive relationships increased prediction accuracy by 5-17% compared to GBLUP.
• Differences between KBLUP and kernel-based SSI were smaller and not always significant.

Conclusions:

• Combining SSI with kernel methods shows potential for enhancing genomic prediction accuracy in maize.
• SSI offers a valuable approach to optimize training data selection for improved genomic predictions.
• Further research is needed to fully elucidate the benefits of kernel-based SSI in diverse breeding programs.