Genomic prediction with multiple biparental families.

Area of Science:

• Plant breeding
• Quantitative genetics
• Genomic prediction

Background:

• Genomic prediction (GP) utilizes genomic and phenotypic data for predicting breeding values.
• In recycling breeding, biparental families (BPFs) are common, generating large datasets for GP.
• Training set composition significantly impacts GP accuracy within BPFs.

Purpose of the Study:

• To evaluate the impact of different training set (TS) compositions on genomic prediction accuracy.
• To determine optimal TS strategies for biparental families in breeding programs.
• To assess the inclusion of related and unrelated families in TS.

Main Methods:

• Applied GP to simulated and experimental biparental families.
• Compared prediction accuracy ([Formula: see text]) using TS of full-sibs, half-sibs, and unrelated families.
• Analyzed combined TS compositions and simulated optimized scenarios.

Main Results:

• Full-sib families yielded highly positive prediction accuracy.
• Half-sib families generally resulted in positive prediction accuracy.
• Unrelated families often showed negative prediction accuracy, reducing overall performance when included in combined TS.

Conclusions:

• Omitting unrelated families and combining full-sib and selected half-sib families in TS is recommended for robust GP.
• Specific mating designs with ~50 genotypes per family balance prediction accuracy and effective population size.
• This strategy supports high short- and long-term selection progress in breeding programs.