Genomic validation software: USA results

Abstract

The European Union requires a validation of genomic estimated breeding values (GEBV) for quality assurance to market the semen of young bulls. Countries such as the United States of America (USA) routinely validate GEBV internally. For practical reasons, there is interest in using standardized methods. The Interbull GEBV Test is conducted with the GEBVtest software and is an alternative to the conventional EBV validation including genomic information. New features recently added to GEBVtest software were a provision to allow countries with small populations a fairer opportunity to pass the test especially for more complex and less heritable traits. A GEBV validation was performed using the newest version of GEBVtest software with different features applied to USA dairy cattle populations. Five breeds and seven traits were tested: milk yield (MIL), fat yield (FAT), protein yield (PRO), somatic cell score (SCS), longevity (DLO) and calving interval (INT) were tested in all five breeds: Holstein (HOL), Jersey (JER), Brown Swiss (BSW), Ayrshire (RDC) and Guernsey (GUE), whereas direct mastitis (MAS) was tested for HOL. Genomic predictions, i.e., GEBV, from August 2022 were used as the full data set whereas GEBV from August 2018 were used as the reduced data set. Results varied due to population size, trait complexity, data ingestion and model differences. The HOL passed the test for all traits, except MAS due to large amounts of new data added and a model change between 2018 and 2022. The validation process for JER behaved as expected for more heritable traits (MIL, FAT, PRO and SCS), but performed poorly for more complex traits (DLO and INT). The analyses were more complicated and resulted in failures for breeds with smaller populations: BSW, RDC and GUE. The failures can be attributed to the complexity of traits, a small number of candidate bulls, and strict parameters within the GEBVtest software. Additionally, for these smaller population breeds and some traits, the parent average presented higher accuracies than GEBV. In summary, USA breeds with larger populations and traits with high heritability resulted in more stable results, whereas USA breeds with smaller populations and more complex traits are hard to validate with tests often failing. The use of TMACE-based genomic evaluations should be used when large data or model changes occur in target traits. Models that include extra regressions could also help to test for other biases and improve accuracies in small populations and/or for complex traits.