Multipoint linkage analysis via Metropolis jumping kernels.

Multipoint linkage analysis is being performed routinely in medical genetic studies to localize disease genes. This likelihood-based method is very computationally intensive. Exact computations are thus formidable for problems with large number of genetic markers and complex pedigrees. This paper proposes a Monte Carlo method to estimate the required likelihoods. The space of multilocus genotypes is sampled using a hybrid algorithm with a mixture of Gibbs samplers and Metropolis jumping kernels. These samples are essentially realizations of a Markov chain, and are distributed approximately according to the conditional genotype distribution given the observed phenotypic data. We present a simulation study with several eight-point analyses to demonstrate the feasibility of the current method.