Unravelling the ancestral regression: 1. comparative performance with known genome-wide relationships

MUNILLA, S.; GARCIA-BACCINO, C.A.; FORNERIS, N.S.; CANTET, R.J.C.

Auckland

Congreso; 11th World Congress on Genetics Applied to Livestock Production; 2018

No corresponde

The ancestral regression model generalizes the covariance structure assumed under the regular animal model by adding a set of extra path coefficients directly connecting an individual?s breeding value to his grandparents? breeding values. The model was developed for performing genomic selection with a practical advantage related to the ease of computation, as the inverse of the genomic relationship matrix can be directly build-up by means of sequential contributions. To test the performance of the model in terms of prediction, we used data from a simulation experiment. The objective of this study was to compare the performance of the ancestral regression model when the true genome-wide relationships between an individual and his four grandparents are known. The model was compared to a regular animal model and we also fitted a single-step genomic selection with the true genomic relationship matrix as a reference. The simulation experiment involved a simplified scenario of a pig nucleus breeding programme for an additive trait controlled by 250 segregating QTLs. After an historical population created to attain mutation-drift equilibrium with a realistic level of linkage disequilibrium, five new generations were developed by mating 20 boars to 200 sows and producing 2,000 offspring. The individuals in the last generation were considered candidates to selection and we focus on predicting their breeding values. Four statistics were computed: accuracy, bias, inflation and mean square error. All models behave very well in terms on bias and inflation. In turn, differences arose in terms of accuracy. The reference model showed a mean accuracy (across replicates) of 0.76, whereas the ancestral regression model outperformed the regular animal model (0.41 vs 0.32), showing that some knowledge was gained when including direct path coefficients from the grandparents.