Improving multi-harvest data analysis in cacao breeding using random regression

This study investigates the application of random regression models for analyzing multi-harvest data in cacao breeding. The aim was to understand the genetic dynamics over ten harvest years and select high-performing genotypes. The trial was conducted in Ouro Preto D’Oeste, Rondônia, Brazilian Amazon. Twenty biparental cacao crosses were evaluated over ten years using random regression models. Models with different polynomial degrees and covariance structures for the residual effects were compared, and the best model was determined using Akaike Information Criterion. We also compared the genetic gains after selecting using three criteria: breeding values, persistence, and area under genotypic trajectories. The best random regression models differed between traits. Genotype-by-harvest interactions were observed, emphasizing the temporal variability in genotype performance. Genetic correlations across harvests illustrated the dynamic nature of genetic expression. Accuracy and heritability fluctuated over successive harvests, emphasizing the complexity of genotype performance prediction. Non-linear genotypic trajectories revealed the presence of unique genetic attributes associated with each trait, with number of healthy fruits showing a tendency towards standardization and dry bean weight displaying a more complex pattern. Consistency in selecting genotypes based on number of healthy fruits highlights reliable selection. Conversely, the variability in choosing top genotypes for dry bean weight underscores the need for cautious selection strategies, as it is a more complex trait to optimize. Despite these insights, future research should consider specific environmental conditions, management practices, and the integration of genomic information for a more comprehensive understanding of genetic dynamics in cacao breeding.