A simple data pruning algorithm to reduce computational load in genetic evaluation of population-traits with high rates of missing phenotypes

Resumen

This study presents a streamlined data pruning algorithm aimed to reducing the computational load in genetic evaluations, particularly for population-traits with high proportions of missing phenotypes. The method addresses the challenge posed by large datasets by selectively retaining only phenotyped animals and their parents for the primary genetic evaluation. This approach, applicable to both pedigree-based (BLUP) and hybrid (ssGBLUP) genetic evaluation methods, significantly reduces the number of equations to solve, thereby reducing computational requirements. The pruning process involves either upward pedigree tracing from phenotyped animals (and genotyped animals in the case of ssGBLUP), or the iterative removal of non-phenotyped non-parents. Genetic evaluations for breeding values and reliabilities are then conducted on the reduced dataset. Subsequently, evaluations for pruned animals are back-solved iteratively, calculating their breeding values and reliabilities based on those of their parents from the main analysis. This iterative back-solving ensures that all animals receive an evaluation. Due to no phenotypic data loss, the estimated and breeding values and reliabilities, and the back-solved breeding values are not compromised. However, back-solved reliabilities for pruned animals were inflated due to the back-solving assumption of unrelated parents. The algorithms and corresponding R code are provided for educational purposes, offering a practical and efficient solution for genetic evaluation in sparse-data contexts.