In collaboration with the Technical University of Delft, researchers at Wageningen University & Research, Animal Breeding and Genomics (WUR-ABG), recently implemented an efficient algorithm, called the deflated preconditioned conjugate gradient method. This algorithm, new in animal breeding, aims to allow large routine single-step genomic evaluations to predict genomic breeding values more accurately. First tests of this new algorithm were performed on large datasets provided by the Breed4Food partner CRV BV, and their results showed promising efficiency.
Large and efficient genomic evaluations
Breed4Food is a consortium established by Wageningen University & Research and four international animal breeding companies: CRV BV, Hendrix Genetics, Topigs Norsvin, and Cobb Europe. These four international companies use (or will use soon) the so-called single-step genomic methods to routinely estimate genomic breeding values of their selection candidates. The single-step methods are appealing due to their simplicity of combining simultaneously traditionally-recorded data with recently-recorded genomic information. Unfortunately, the fast increase of genomic information limits the feasibility of routine single-step genomic evaluations with software and algorithms currently used.
A new algorithm in animal breeding
Within Breed4Food, and in collaboration with Prof. Kees Vuik of the Technical University of Delft, researchers at WUR-ABG have integrated in their software (under development), an efficient algorithm, called deflated preconditioned conjugate gradient method. This algorithm allows to perform large single-step genomic evaluations efficiently. This algorithm has never been used in animal breeding, while it has shown good results in other fields, such as in bubbly flows problems.
First results are promising
First tests of this new algorithm were performed on large datasets provided by the Breed4Food partner CRV BV. These results showed promising efficiency of the new algorithm, which will allow to perform routinely large single-step genomic evaluations. Furthermore, it has been shown that this new algorithm could be easily implemented in software currently used in animal breeding.