Integrating genomic additive relationship matrices improves the efficiency in diploid banana breeding

Partitioning of genetic variance into additive and non-additive components using the pedigree-based best linear unbiased prediction (P-BLUP) model is possible because of the family structure and replicated clones in clonally propagated crops, but this model may overestimate these components. However, the genomic best linear unbiased prediction (G-BLUP) method, which integrates the genetic relationship through molecular marker information reduces the overestimation. Alternatively, a combination of the P-BLUP and G-BLUP, sourcing to create a hybrid matrix that estimates hybrid best linear unbiased prediction (H-BLUP), is proposed. We investigated if integrating molecular information into the clonal model could improve the partitioning of the variance components leading to more accurate estimates of genetic parameters and prediction accuracy of breeding values of 14 key traits in diploid banana. In this study, we used clones of 14 full-sib families from a factorial mating design of four female and five diploid male banana (Musa acuminata) parents, generated at the International Institute of Tropical Agriculture in Arusha. The genomic-based relationship matrices were constructed using a set of 2792 filtered single-nucleotide polymorphism markers. Additive variance and heritability derived from G-BLUP and H-BLUP models reduced bias compared to the P-BLUP model. The H-BLUP estimated the highest prediction accuracies for yield-related and cycling traits, while the P-BLUP model had the highest prediction accuracy estimates for agronomic traits. The use of marker-based models enhances the accuracy of predicting breeding values, contributing to accurate estimates of genetic gain while paving a way for further genomic exploration in diploid banana breeding programs.