Influence of phenotyping truncation on genetic parameter estimate and genomic prediction of productive traits in nellore cattle

Abstract

This study evaluated the impact of progressively reducing phenotypic data collection on variance components and the genomic prediction ability of productive traits in Nellore cattle. A total of 288,456 phenotypic records and 51,471 genotypes were obtained from the National Association of Breeders and Researchers (ANCP), covering three traits: rib eye area (REA), residual feed intake (RFI), and adjusted weight at 450 days (W450). Two data-reduction scenarios were simulated: a retrospective scenario, involving the sequential removal of the most recent data by year, and a prospective scenario, involving the annual exclusion of historical data. In both scenarios, animals born in 2021 comprised the validation set. Analyses were performed using the single-step genomic best linear unbiased prediction method (ssGBLUP), with prior estimation of variance components adjusted for each scenario and year. Prediction accuracy, bias, and dispersion were evaluated using linear regression (LR) between genomic estimated breeding values (GEBV) from the complete and reduced datasets. In the retrospective scenario, starting from the complete 2021 database and removing contemporary records down to 2013, prediction accuracy declined from 0.72 to 0.299 for REA, from 0.51 to 0.11 for RFI, and from 0.77 to 0.34 for W450, accompanied by reduced heritability estimates and increased underdispersion. These effects were mainly attributed to the loss of genomic connectivity and the low proportion of genotyped animals (<1% in recent years). For RFI, genetic variance also increased due to the concentration of data on farms with strong genomic structure, although the absence of contemporary records limited predictive ability. Conversely, in the prospective scenario, starting from 2013 dataset and progressively removing the recent records through 2021, accuracy remained essentially stable, changing from 0.72 to 0.73 for REA, from 0.51 to 0.50 for RFI, and from 0.76 to 0.77 for W450, while bias and dispersion stayed within acceptable limits. This outcome was supported by the presence of up-to-date phenotypes and a higher proportion of genotyped animals (exceeding 70% by 2018). These findings demonstrate that while the exclusion of historical data can be a viable strategy to reduce computational demands, maintaining a contemporary and representative phenotypic dataset is critical to ensuring the reliability of genomic predictions and the long-term success of breeding programs.