A mixture model approach is employed for the mapping of quantitative
trait loci (QTL) for the
situation where individuals, in an outbred population, are selectively
genotyped. Maximum
likelihood estimation of model parameters is obtained from an Expectation-Maximization
(EM)
algorithm facilitated by Monte Carlo sampling using a Gibbs sampler. All
individuals with
phenotypes, whether genotyped or not, are included in the analysis where
both putative QTLs and
missing marker genotypes are sampled conditional on known marker information
and phenotype.
A simulation of a half-sib family structure demonstrates that this mixture
model approach will
yield unbiased estimates of the allelic effects of a QTL affecting the
trait on which selective
genotyping is based. Unbiased estimates were also obtained for the QTL
effect on a correlated
trait provided both traits were analysed jointly in a bivariate model.
The procedure is also
compared with a standard linear model approach. The application of these
methods is
demonstrated for bovine chromosome six, the data arising
from two Holstein–Friesian families
selectively genotyped for protein yield in a daughter design.