Study On Single-step Strategy Based On Random Regression Model In Genomic Selection

Genomic selection has been widely implemented in genetic evaluations of domestic animals.It is a form of maker-assisted selection using markers covering the whole genome.Methods of genomic selection can be classified into multi-step methods and single-step methods.In genetic evaluations of traits with a single record,single-step methods have the advantages of simplicity of implementation and less bias of prediction compared with multi-step methods.However,in genetic evaluations of longitudinal traits with multiple records,the benefits of single-step method had not been explored.The goal of our study was to establish the time series model for genomic selection of longitudinal traits,to investigate its performance via validation studies,and further explore its application in genetic evaluations.(1)Following theoretic analyses on the methodology system of genomic selection and traditional genetic evaluation of longitudinal traits,we proposed an approach for genomic evaluation of longitudinal traits,i.e.,single-step random regression model.(2)To evaluate the benefits of single-step random regression model,we performed both simulation study and empirical study.In simulation study,program used for simulation of longitudinal traits was wrote and then data sets were generated under various scenarios.Results showed that single-step random regression model outperformed pedigree-based random regression model and GBLUP in both accuracy and unbiasedness in all scenarios.GAW18 real dataset was used in empirical validation study.Results of cross-validation showed that single-step random regression model achieved higher accuracy than GBLUP.(3)We further investigated the application of single-step random regression model to genomic evaluation of Chinese Holstein population.Firstly,based on traditional multiple-trait multiple-lactation random regression model,we estimated the variance components of milk,fat and protein yield in the first three parities of Chinese Holstein population.With the availability of variance component estimates,we implemented single-step random regression model in genomic evaluation of milk,fat and protein yiled in Chinese Holstein population in Beijing and Shanghai.Results showed that predictions of single-step random regression model were more accurate and less biased than GBLUP.Moreover,parameters in the construction of hybrid relationship matrix had effects on the convergence of iteration and performance of single-step random regression model.Therefore,we should explore the parameters resulting the best performance in applications.In our study,non-commercial program for genetic evaluation named PI-BLUP was wrote,which was among the first to perform large-scale genetic evaluation in China.Single-step random regression model was demonstrated to have better performance and is simpler to be implemented.Single-step random regression model provides a better alternative approach for genomic evaluations of longitudinal traits.The software PI-BLUP could be used in genetic evaluation based on single-step random regression model and other methods.