Heritability estimation with unknown pedigree based on the joint distribution of marker and phenotypic data using Markov chain Monte Carlo techniques
Abstract
The heritability of a quantitative trait is a very important parameter to quantify the genetic variation present in a population. Although traditional techniques for estimating heritability require accurate information of the genetic relationship among individuals, pedigree structure is generally lacking in natural population. Nowadays, the development of DNA markers is making possible to reconstruct pedigree accurately with sufficient markers.
These reconstructed pedigrees have then been used with restricted maximum likelihood under a general linear mixed model to estimate heritability.
In this thesis, we use markers and phenotypic observations jointly to estimate the pedigree and heritability simultaneously. We develop a MCMC sampling method of moving through the sibship configuration space and space of parameters of quantitative trait, and finding the configuration and optimal parameter values that maximizes the full joint likelihood or posterior distribution of proposed family structure and genetic variance components. Using this method, we estimate of heritabilities of 318 abalone at different time points separately and independently. Both MLE and Bayes estimate are superior to two-step method using insufficient markers (two microsatellite markers). We also give the discussion about the choices of prior distributions of parameters in the model. At the end, we extend our method to incorporate with observations at multiple time points, but we don't obtain any significant improvements.