34
Genomic prediction using a model based on haplotype clusters
Marker based models used for genomic prediction are often based on the underlying assumption that the genetic markers are the casual variants. An alternative two-stage model for genomic prediction uses haplotype clusters as hidden states in a hidden Markov model. The genomic effects are modeled as haplotype effects associated with the haplotype clusters instead of marker effects associated with markers. In the first stage, hidden Markov model parameters of the haplotype cluster model are estimated based on the genotypic data. In the second stage, a Bayesian model based on the estimated haplotype cluster model is used to model the underlying haplotype effects at the haplotype loci. The haplotype effect alleles are treated as non-emitting states in the haplotype cluster model. The prior for the haplotype effects is a mixture of a multivariate normal distribution and a point mass at zero distribution. Including a missing genotype as a possible observed state in the hidden Markov model allows for animals to be genotyped with different SNP panels.
Keywords:
Bayesian, GWAS, genomic prediction