Optimizing multi-breed joint genomic prediction issues in numerically small breeds for sex-limited trait in a loosely structured dairy cattle breeding system.

Abstract

Genomic prediction is crucial in the developed dairy industry, but implementing it in resource-poor regions with numerically small breeds and with no historic pedigree information is challenging. This study explores possibilities for joint genomic prediction, using genomic best linear unbiased prediction (GBLUP) across four closely related breeds for sex-limited traits when recently collected genomic information and phenotypes are available. The data was simulated to cover low (0.1) and moderate (0.3) heritability scenarios. Principal Component Analysis (PCA) revealed genetic relatedness among breeds, with the first two components explaining 80% of variance. Combining breeds for genetic evaluation using only genomic information enhanced prediction accuracy and reduced bias in genomically estimated breeding values (GEBV) compared to single-breed models. Ancestry-specific allele frequencies and allelic effects had minimal impact due to genetic similarity between breeds. Multi-breed evaluation substantially improved accuracy. The multi-breed two-tailed selective genotyping model (MTB) had better accuracy of prediction than top-selected (MTOP) and randomly selected (MRND) models. However, looking into standard error for accuracy of prediction of GEBV and least bias of prediction, MRND model is recommended for multi-breed joint prediction evaluation in numerically small breeds. For 0.3 h² scenario, MTOP gained 17.89% accuracy, MTB gained 20%, and MRND gained 24.39% over single breed models. Similar trends were seen in the low heritability (0.1) scenario. For small breeds without pedigree records data, adopting a multi-breed joint evaluation with random selective genotyping is recommended. This strategy has potential to integrate crucial breeds into genomic selection while conserving resources in genotyping and data recording in resource-poor regions.