Accurate Prediction of Genomic Breeding Values Across Families Combining Linkage Disequilibrium and Co-Segregation30 April 2013
Prediction inaccuracies of linkage disequilibrium are a problem to geneticists, however a team of researchers has looked into a 'linear mixed model' with the benefit of greater accuracy.
Traditional genomic prediction methods relying only on linkage disequilibrium (LD) between marker and QTL can have low accuracy because LD is not likely to be consistent in phase or strength across different families.
A linear mixed model fitting both genome-wide co-segregation (CS) and LD (LD-CS model) is developed to improve prediction accuracy across families.
CS is modeled as the effects of all 1-centimorgan haplotypes that one individual inherits from pedigree founders through identity-by-descent, while LD is modeled as allele substitution effects for all marker genotypes of that individual.
Prediction accuracy of LD-CS model was compared to three LD methods—GBLUP, BayesA and BayesB, using simulated datasets of paternal half sib families varying in number of sires. Within each sire family, 10 half sibs were used for training and to predict breeding values for another 10 half sibs.
All individuals had phenotypes for a quantitative trait with heritability 0.5 and genotypes for 2,000 SNPs. Results showed that LD-CS model had significantly higher accuracy than any LD methods except for the dataset of one sire family.
With the increase in the number of families, the accuracies of LD-CS model persisted while those of LD methods dropped. In conclusion, by fitting CS explicitly the LD-CS model potentially have higher and more consistent prediction accuracy across families than LD methods.
Table 1. Prediction Accuracy of LD-CS Model and LD Model Using GBLUP, Bayesa And BayesB for Simulated Datasets with Different Number of Sires (No. Sires). Results are the Average of 32 Replicates For Each Number Of Sires.