# Combining ability of banana triploid hybrid progenitors and genomic prediction of cross performance for agro-morphological traits

## TL;DR

This study explores how to breed better bananas by analyzing the genetic contributions of parents and predicting hybrid performance using genomic data.

## Contribution

The study introduces genomic prediction methods for cross performance in banana breeding using combining ability and epistasis.

## Key findings

- Heritability was moderate to high, with tetraploid parents predominantly influencing hybrid performance.
- Genomic prediction achieved predictive abilities over 0.5, especially when tetraploid parents were shared across crosses.
- Combining ability and epistasis analysis can guide parent selection for improved banana hybrids.

## Abstract

Breeding disease-resistant cultivars that meet commercial criteria is essential to sustain banana production threatened by major diseases. Edible bananas are seedless triploid hybrids that represent end-breeding products. Hence, the crucial step in banana breeding is to improve and combine the parents. Currently, little information is available on parental combining abilities and on the inheritance of major traits to effectively guide banana breeding strategies. In this study, a breeding population of 2,723 triploid individuals resulting from multiparental diploid-tetraploid crosses was characterized during three crop cycles for 23 traits relating to plant and fruit architecture and bunch yield components. The phenotypic variance was partitioned between non-genetic and genetic effects, the latter including the general combining ability of diploid and tetraploid parents, their specific combining ability, and additional variance due to the within-cross genetic variability. Heritability was moderate to high depending on the trait and revealed the predominance of the tetraploid parent's contribution to hybrid performance for most traits. The use of parental genomic information enabled cross-mean performance prediction through genomic relationship matrices of general and specific combining abilities, the latter being partitioned into dominance and across-population epistasis contributions. Predictive abilities often greater than 0.5 were obtained, particularly when the tetraploid parent was observed in other crosses and, for some traits, when neither parent was observed. Information on trait inheritance and genomic prediction of cross-mean performance will help in selecting and combining parents, facilitating the identification of promising hybrids.