Genomic prediction within and across maize landrace derived populations using haplotypes

Area of Science:

• Plant breeding and genetics
• Quantitative genetics
• Bioinformatics

Background:

• Genomic prediction (GP) leverages high-density markers like single nucleotide polymorphisms (SNPs) for breeding value estimation.
• Haplotype-based GP is theoretically superior to SNP-based GP due to better capture of ancestral information and linkage disequilibrium with quantitative trait loci (QTL).
• However, empirical evidence for haplotype-based GP advantages is variable, influenced by population structure, traits, and haplotype construction methods.

Purpose of the Study:

• To compare the prediction accuracies of haplotype-based versus SNP-based genomic prediction in diverse maize populations.
• To evaluate the impact of three different haplotype construction methods (FixedHB, HaploView, HaploBlocker) on prediction performance.
• To investigate prediction scenarios within and across populations derived from European maize landraces.

Main Methods:

• Generated and genotyped four populations (doubled haploid and gamete capture lines) from two European maize landraces with 600k SNPs.
• Phenotyped lines for five traits and performed genomic prediction using SNP data and three distinct haplotype construction methods.
• Evaluated prediction accuracies in within-population, across-population (DH vs. GC), and across-landrace scenarios.

Main Results:

• Within-population predictions: FixedHB and HaploView methods performed comparably or slightly better than SNPs; HaploBlocker showed variable results.
• Across-population predictions: HaploBlocker's performance was highly dependent on parameter settings for shared haplotype construction.
• Across-landrace predictions: Both SNP and haplotype approaches yielded low accuracies, though HaploBlocker offered substantial improvements for specific traits.

Conclusions:

• Haplotype construction method choice significantly impacts GP accuracy, especially in cross-population predictions.
• HaploBlocker demonstrates potential for improving genomic prediction across populations, particularly when optimized for shared haplotype detection.
• The study provides insights into optimal haplotype construction strategies and parameter selection for effective genomic prediction in maize breeding programs.