295. Benchmark SNPs selection strategies to achieve optimum prediction in small population

Abstract

In recent decades, genomic selection has become a routine practice for genetic evaluations of most livestock species and breeds. However, its use is still limited in small local populations. The main reasons are the lower interest to invest in these breeds, and the poor accuracy of their genomic predictions due to the low number of animals. Bayesian variable selection regression models (i.e. Bayes B/C) appear to improve the accuracy of genomic prediction from small populations, but their ability to account for selection remains unclear, especially when using time-series cross validations. In the present work, we evaluated in terms of accuracy and bias the impact of constructing the genomic matrix using only a subset of informative SNPs. Analysis was performed on Rendena cattle, a small local cattle population of North Italy.