Lee Deukhwan, Misztal Ignacy, Bertrand J Keith, Rekaya Romdhane
Department of Animal and Dairy Science, University of Georgia, Athens, GA 30605, USA.
J Appl Genet. 2002;43(2):209-16.
Data included 393,097 calving ease, 129,520 gestation length, and 412,484 birth weight records on 412,484 Gelbvieh cattle. Additionally, pedigrees were available on 72,123 animals. Included in the models were effects of sex and age of dam, treated as fixed, as well as direct, maternal genetic and permanent environmental effects and effects of contemporary group (herd-year-season), treated as random. In all analyses, birth weight and gestation length were treated as continuous traits. Calving ease (CE) was treated either as a continuous trait in a mixed linear model (LM), or as a categorical trait in linear-threshold models (LTM). Solutions in TM obtained by empirical Bayes (TMEB) and Monte Carlo (TMMC) methodologies were compared with those by LM. Due to the computational cost, only 10,000 samples were obtained for TMMC. For calving ease, correlations between LM and TMEB were 0.86 and 0.78 for direct and maternal genetic effects, respectively. The same correlations but between TMEB and TMMC were 1.00 and 0.98, respectively. The correlations between LM and TMMC were 0.85 and 0.75, respectively. The correlations for the linear traits were above.97 between LM and TMEB but as low as 0.91 between LM and TMMC, suggesting insufficient convergence of TMMC. Computing time required was about 2 hrs, 5 hrs, and 6 days for LM, TMEB and TMMC, respectively, and memory requirements were 169, 171, and 445 megabytes, respectively. Bayesian implementation of threshold model is simple, can be extended to multiple categorical traits, and allows easy calculation of accuracies; however, computing time is prohibitively long for large models.
数据包括412,484头盖尔维牛的393,097条产犊难易度记录、129,520条妊娠期长度记录和412,484条出生体重记录。此外，有72,123头动物的系谱信息。模型中纳入了作为固定效应的母畜性别和年龄效应，以及作为随机效应的直接遗传效应、母体遗传效应、永久环境效应和当代组（畜群 - 年份 - 季节）效应。在所有分析中，出生体重和妊娠期长度被视为连续性状。产犊难易度（CE）在混合线性模型（LM）中被视为连续性状，或在线性阈值模型（LTM）中被视为分类性状。将通过经验贝叶斯（TMEB）和蒙特卡罗（TMMC）方法在TM中获得的解与LM的解进行比较。由于计算成本，TMMC仅获得10,000个样本。对于产犊难易度，LM与TMEB之间直接遗传效应和母体遗传效应的相关性分别为0.86和0.78。TMEB与TMMC之间相同效应的相关性分别为1.00和0.98。LM与TMMC之间的相关性分别为0.85和0.75。线性性状在LM与TMEB之间的相关性高于0.97，但在LM与TMMC之间低至0.91，表明TMMC收敛不足。LM、TMEB和TMMC所需的计算时间分别约为2小时、5小时和6天，内存需求分别为169、171和445兆字节。阈值模型的贝叶斯实现简单，可以扩展到多个分类性状，并允许轻松计算准确性；然而，对于大型模型，计算时间长得令人望而却步。
