基于JMP软件的Lasso及岭回归在水稻全基因组预测中的应用

Application of Lasso and ridge regression in rice whole-genome prediction based on JMP software

全基因组选择是21世纪动植物育种的一种重要的选择策略,其核心就是全基因组预测,即基于分布在整个基因组上的多样性分子标记来对育种值进行预测,为个体的选择提供依据.但目前提出的大多数全基因组预测方法都涉及到相当复杂的算法并要求使用者具备熟练的编程能力,因此很少在实际育种中得到有效的应用.对统计软件JMP在水稻全基因组预测中的应用做了探索研究,利用其中的两种正则化回归方法(Lasso和岭回归)预测产量及其相关性状的育种值,结果表明两种方法对这些性状均有较好的预测效果,但Lasso优于岭回归,预测效果因性状不同也有差异,4个性状预测效果的次序为:千粒重>分蘖数>单株谷粒数>产量.鉴于JMP软件的强大的分析能力、友好的用户界面和可操作性,本研究的结果可以为育种工作者在选择应用全基因组预测的分析工具方面提供较好的参考.

Whole-genome selection is an important selection strategy for animal and plant breeding in the 21st century.The core of the whole-genome selection is whole-genome prediction,that is,the breeding value is predicted based on the diversity of molecular markers distributed throughout the genome,providing a basis for individual selection.However,most of the genome-wide prediction methods currently proposed involve fairly complex algorithms and require users to have proficient programming skills,so they have rarely been effectively applied in actual breeding.This paper explored the application of the statistical software JMP in rice genome prediction,and usesd two of the regularization regression methods(Lasso and ridge regression)to predict the breeding value of yield and related traits.The results show that both methods have good prediction effects,but Lasso is better than the ridge regression.The prediction effects also vary depending on the traits.The order of the 4 trait prediction effects is:kgw>tiller>grain>yield.In view of the powerful analysis capabilities,friendly user interface,and operability of the JMP software,the results of our research can provide a better reference for breeders in selecting analysis tools for genome-wide prediction.