Haplotype blocks for genomic prediction: a comparative evaluation in multiple crop datasets.

Area of Science:

• Plant Breeding and Genetics
• Genomics and Bioinformatics

Background:

• Genomic selection is crucial in modern plant breeding for identifying superior genotypes.
• High marker density in breeding populations leads to redundant genotype data, prompting interest in haplotype blocks.
• Haplotype blocks can summarize co-inherited features and capture local epistasis.

Purpose of the Study:

• To compare the prediction accuracy of genomic selection methods using single nucleotide polymorphisms (SNPs) versus haplotype blocks.
• To evaluate different methods for constructing haplotype blocks and their impact on prediction accuracy across various crop datasets.

Main Methods:

• Utilized four public crop datasets (canola, maize, wheat, soybean).
• Compared prediction accuracies of methods using SNPs and haplotype blocks.
• Investigated various haplotype block construction approaches (LD-based, physical distance, marker count, Haploview, HaploBlocker) and prediction models (GBLUP, EGBLUP, Bayesian LASSO, RKHS).

Main Results:

• Haplotype blocks improved prediction accuracy for some traits compared to SNP-based predictions, with improvements being trait- and model-specific.
• Haplotype blocks particularly enhanced accuracy in low marker density scenarios.
• Physically large haplotype blocks generally decreased prediction accuracy; no single method for block construction proved universally superior.

Conclusions:

• Haplotype blocks offer potential for improving genomic prediction, especially for underperforming models or in low marker density situations.
• The optimal method for defining haplotype blocks is dataset- and trait-dependent, requiring adjustment as hyperparameters.
• Flexible criteria for haplotype block construction are necessary for maximizing genomic prediction accuracy in plant breeding.