495
Combining SNPs in latent variables to improve genomic prediction
The objective of this study was to develop and test hierarchical genomic models with latent variables that represent parts of the genomic values. An interaction model and a chromosome model were compared with a model based on variable selection in a simulated and real dataset. The program Bayz was used to calculate the parameters which were subsequently used to predict breeding value or the pre-corrected phenotypes in a cross validation.
The predictive value did not vary much for the simulated dataset among models and was in line with earlier results. Correlations between predicted and true breeding were around 0.9. For the mice dataset cross validation correlations were around 0.5. Using latent vectors to combine snps in genomic prediction models allows for estimation of non-linear effects such as interaction among snps and the use of prior biological information regarding the snps.
Keywords:
Hierarchical genetic model
Predictive value
Gibbs sampling
