Multipoint linkage analysis of complex traits with Markov chain Monte Carlo linkage analysis. G. Snow1, E. Wijsman1, E. Thompson1, S. Heath2. 1) Biostatistics, University of Washington, Seattle, WA; 2) Rockefeller University, New York, NY.
Lod scores for multipoint linkage analysis can be difficult or impossible to compute for large and/or complex pedigrees when there are many multiallelic markers. Markov chain Monte Carlo (MCMC) methods offer one computationally tractable solution. MCMC methods are based on realization of dependent samples of genotypes, from which desired parameters and likelihood ratios can be estimated. Earlier MCMC samplers for lod score computations are based on single-genotype updating (e.g., the Gibbs sampler), which tends to be relatively inefficient thus limiting practical utility.
Here we show that use of a sampler which simultaneously updates all genotypes at each locus in turn is much more efficient. We also incorporate a mixed-model for the trait locus, thus allowing for a more complex model. We applied this approach to a 101-member simple-structure pedigree with simulated trait and marker data, and compare results to those obtained by Vitesse by combining environmental and polygenic variance. For 4 markers spaced at 20 cM intervals, a linked quantitative trait with major gene heritability .69 and polygenic heritability .15, the peak lod score (5.6) for the MCMC analysis is virtually identical to that obtained by Vitesse (5.7), and both peaks occur at the correct location. However, the MCMC approach provides more accurate localization (1 lod support interval ~50% that provided by analysis with Vitesse) and requires less CPU time. When 10% of the pedigree is unobserved, for 4 markers Vitesse requires 115 minutes vs.20 minutes for the MCMC approach; for 5 markers, Vitesse requires >12 hours, while < 30 minutes are required for the MCMC approach. This demonstrates the practical utility of this approach for multipoint linkage analysis of large pedigrees.
Supported by NIH GM 46255
.