Statistical Analysis And Software Development Of Genetic Model For Half-sib Progeny Test In Forest Trees

Progeny test is a key point in forest tree genetic improvement, which affects the process of forest tree breeding programs directly. Based on the progeny test, we can estimate genetic parameters such as heritability to provide a reliable basis for the evaluation and selection of breeding material, and hence promote the process of forest tree genetic improvement. But there still exist some problems in half-sib progeny test in calculating the genetic parameters and some statistics, especially when the experimental data are imbalanced. Although some statistic software such as SAS and ASReml can be used to calculate genetic parameters, it is quite difficult for many breeders to use them. The purpose of this study is to clarify the statistical theories and methods related to the calculation of genetic parameters in forest tree progeny test, and then create R packages to provide a simple and effective tool for the calculation of genetic parameters.For the balanced and imbalanced experimental data from half-sib progeny test at multiple site, firstly, a fixed-effect model was established and the least square method was employed to calculating the estimates of general combining abilities and other fixed effects. Secondly, a random-effect model was built and the analysis of variance method was used to analyze the genetic variance components and then calculate the heritability. Meanwhile, the formulae were derived for calculating standard errors of the estimators and the statistic for hypothesis test of genetic variance components. For multiple quantitative traits, the methods of calculating the genetic correlation coefficient and its standard error between any two traits at the levels of family, site×family and block(site)×family effects were presented. Then, the mixed model equation was constructed and the method of REML was applied to estimate the genetic variance components and genetic covariance and then analyze the heritability and genetic correlation coefficient. Furthermore, the method of using the mixed model equation to obtain the best linear unbiased prediction of breeding values for families and their offspring was elaborated. We demonstrated that the simple model of family breeding values can be used to calculate the breeding values for large-scale offspring individuals. Finally, R packages named Halfsib MS and Halfsib BV were developed based on the method of MLS and ANOVA, and the method of REML and BLUP for the genetic model of half-sib progeny test in forest trees, respectively.The program packages are of simple operation and need no programming. Furthermore, the output of the programs is comprehensive and easy to understand. Users can freely download the packages from the website http://fgbio.njfu.edu.cn/tong/halfsibms/halfsibms.html and http://fgbio.njfu.edu.cn/tong/halfsibbv/halfsibbv.html.