31 Changes in genomic predictions when new data is included

Abstract

Listen

Translate article

Abstract Persistence of genomic evaluations depends on the amount of data and population parameters. When the data are large enough to estimate nearly all chromosome segments (about 10,000 in Angus), the persistence can be high. Subsequently, the regular evaluations can be run less often, and indirect predictions based on backsolved SNP effects may be accurate over an extended period as full predictions. The objective of this study was to investigate changes in GEBV across monthly genomic evaluations in a large genotyped population. A dataset from American Angus that contained 8,186,503 records for birth weight, 8,881,124 for weaning weight, and 4,386,184 for post-weaning gain was used. A total of 10,129,980 animals born up to December of 2017 had pedigree information, of which 484,074 were genotyped. A truncated dataset kept animals that were born up to December of 2016 (Dec2016). To mimic a scenario where evaluations are run every 30 days, the 2017 data was added month by month and new GEBV were computed using ssGBLUP. Genomic predictions from the Dec2016 evaluation were contrasted to the monthly evaluations for all genotyped animals, genotyped animals with or without phenotypes, genotyped animals with or without new phenotypes after Dec2016 and for young genotyped animals (born in 2016, with no phenotypes or progeny). The average number of animals added by month was 26,225, with a larger concentration from January to April. The correlations between GEBV from Dec2016 and all the 12 consecutive evaluations were greater or equal than 0.99 in all cases. The average (maximum) changes for young animals from Dec2016 to January and December of 2017 ranged from 0.05 (0.25) to 0.10 (0.53) additive standard deviations. The same changes for animals with records past Dec2016 were from 0.05 (0.11) to 0.88 (1.59). Genomic evaluations have high persistence over a year; however, outliers show larger changes.