Age dependent QTL analysis using Gibbs sampling for random effects models

Repeated measurements of traits contain more information than a single cross-sectional measurement for inference on age-dependent genetic effects in humans. However, most existing linkage analysis approaches either use only derived summary measures from the repeated measurements or are only applicable to very limited patterns of age-dependent genetic effects. This thesis provides a general extended variance components approach using Markov Chain Monte Carlo (MCMC) methods to fully utilize data collected longitudinally in extended human families to calculate age-dependent heritabilities for quantitative trait loci (QTL) as well as other parameters of interest.; We first developed the model and methodology assuming that we have complete data on all members of the extended families. This model allowed for a polynomial representation of the QTL genetic effect, but limited the residual genetic effect to a scalar. We examined the behavior of this approach under various priors and sample sizes. For each scenario we used a burn in of 10,000 iterations. Our estimates for the distributions of the parameters are drawn from every 50th iteration of the last 30,000 iterations. We extended these strategies to situations where phenotype data could be missing on some individuals in the pedigrees. Assuming an ignorable missing mechanism, data augmentation methods were applied to incorporate missing data. Finally, we used the Deviance Information Criterion for model selection.; We found that non-informative priors and large sample sizes (3000) yielded the best resulting distributions of the parameters. In addition to simulation studies, we applied our method to high-density lipoprotein (HDL) data collected in the Framingham Heart Study. This application used data collected from 1971 to 1996 in the Framingham Offspring Study and genotype data for the marker GATA184A08 on chromosome 6. The Deviance Information Criteria was used for model comparisons. A quadratic age dependent model provided the best fit among the models we considered. In these data there is clear evidence that the quantitative trait locus effect on chromosome 6 is increasing among people aged 40 to 70. We conclude that these procedures may provide a useful tool for evaluating age-dependent genetic effects in data collected longitudinally.