Submitted to: American Society of Animal Science Annual Meeting
Publication Type: Abstract Only
Publication Acceptance Date: 3/8/2007
Publication Date: 7/10/2007
Citation: Kuehn, L.A., Thallman, R.M., Leymaster, K.A. 2007. Evaluating the feasibility of fitting haplotype effects as random: Variance component estimation [abstract]. Journal of Animal Science. 85(Supplement 1):669-670. Abstract #945.
Technical Abstract: Fitting haplotypes as random effects in association studies may prevent overestimation of haplotypic effects with low frequencies. The objective was to determine whether haplotypic variance could be accurately estimated. Using simulation, haplotypic effects were deterministically assigned to either 2 or 16 haplotypes with variances of haplotypic effects set at 2, 4, or 16.67 units**2. Haplotypes were assigned stochastically to base animals in frequencies such that the 3 haplotypic variance scenarios accounted for 3, 6, and 25 units**2 of genetic variance, respectively. Polygenic additive effects were assigned from a normal distribution so that a total of 25 units**2 of the phenotypic variance was genetic. Residual effects were sampled from a normal distribution with a variance of 75 units**2. Either 250 or 1,000 progeny (5 or 15 per sire) with performance records were simulated. With varying levels of haplotype number, haplotypic variance, numbers of progeny, and numbers of progeny per sire, there were a total of 24 different simulation scenarios. Each simulation scenario was replicated 100 times and analyzed using MTDFREML with a model that included a random independent regression for the number of copies of each haplotype (0, 1, or 2), a random polygenic effect, and error. Estimates of polygenic and residual variance were accurate for all scenarios. Standard errors of the mean estimates for both variance components were greater with 250 than 1,000 progeny and with 5 than 15 progeny per sire. When 16 haplotypes were simulated, the variance of haplotypic effects were accurately estimated when the pedigree contained 1,000 progeny and underestimated with 250 progeny. Haplotypic variance was consistently overestimated when only 2 haplotypes were simulated, possibly due to the small number of classes for the haplotypic regression. Large, phenotyped pedigrees are important when estimating haplotypic variance for association studies.