利用广义线性模型预测家畜离散性状育种值  

Predicting Breeding Values for Discrete Traits Using Generalized Linear Mixed Models

作　　者：殷宗俊[1] 张勤[2] 

机构地区：[1]安徽农业大学动物科学系,安徽合肥230036 [2]中国农业大学动物科学系,北京100094

出　　处：《中国畜牧杂志》2006年第11期1-4,共4页Chinese Journal of Animal Science

基　　金：国家重点基础研究发展规划(批准号:G2000016103);安徽省自然科学基金(050410204);安徽省教育厅项目(批准号:2002jq126;2004kj151)

摘　　要：将广义线性混合模型(GLMM)引入动物离散性状的遗传分析及个体的遗传评定,初步比较了GLMM方法与一般线性方法(LM)的估计效果。模拟研究的性状为单阈值二项分类性状,选用的连接函数为对数连接μi=eη/(1+eη),方差函数为V(μi)=μ(i1-μi)/n,试验设计为全同胞-半同胞混合家系,参数估计采用Fisher迹法。结果表明:GLMM方法能较准确地估计公畜的个体育种值,在个体的遗传评定效果方面要明显优于常规的线性方法,其预测的育种值排序结果与真实育种值的排序之间存在极显著的相关性(P<0.001)。A method of generalized linear mixed model （GLMM） was introduced for genetic analysis and breeding value prediction for discrete traits. The logit linkage function （μi = e^η/（1 ＋ e^η） was applied. The method of Fisher Scoring was used to solve the non-linear equations. To exam the efficiency of the proposed method and to compare the conventional linear model （LM） method, a simulation study was conducted, in which a binary trait and a population with multiple full and half sib families were simulated. The results showed that the GLMM method had a great advantage in predicting breeding values for discrete traits. The Sperman correlation between the predicted and the true breeding values was 0.926 for GLMM method, which is much higher than that from LM method （0.665）.

关 键 词：畜牧学 广义线性混合模型 离散性状 估计育种值 家畜 

分 类 号：S813.2[农业科学—畜牧学]