Application Of Generalized Regression Model In Rice Genomic Prediction Using JMP Software

Simulation and prediction before breeding through biotechnology can greatly save manpower and material resources and provide a good theoretical basis for subsequent agricultural production.The basis for molecular marker breeding in whole genome selection is to first calculate the estimated breeding value of the whole genome,and the calculation of the estimated breeding value of the whole genome requires us to understand all the markers or haplotype effects in the whole genome.At the same time,using the whole genome to predict rice agronomic traits can improve the accuracy of breeding value estimation,reduce costs,facilitate early selection,and accelerate genetic progress.The key problems of whole-genome prediction breeding are mainly concentrated in two aspects.First,choosing appropriate and concise statistical software can effectively solve the problem of variable selection,eliminating useless data,and ensuring that the data fitted by the model is useful.Second,for the predicted crop traits,a suitable regression model can be selected in the software.In this paper,we explore and discuss these two main issues.In this paper,several generalized regression models in the software JMP are used to make a genome-wide prediction of two data materials of rice.The two data sets are the two-year yield,flowering period and plant height of the recombinant inbred line population of 369 lines.Genome-wide prediction and yield of 210 inbred recombinant inbred line populations and their related traits: thousandgrain weight,number of tillers and number of grains per plant were used for genomewide prediction,and the parents of both data sets were Zhenshan 97 and Minghui 63.We explored the effect of each model in JMP on the whole genome prediction of rice traits through the above data materials.We also explored the predictive variable screening function in JMP,and compared the models for predicting each agronomic trait of rice,and selected the one suitable for each trait.For the better model,the main results obtained are as follows:1.Using five generalized regression models Lasso,elastic network,ridge regression,adaptive Lasso and adaptive elastic network in the JMP to rice materials for369 strains to perform genome-wide prediction to obtain good results,the predicted and actual values correlation coefficient is between 0.6 and 0.9.Choose 369 strains and73147 SNP sites.This is a relatively large data.We want to explore the function of variable selection in JMP to eliminate invalid data.For the problem of p > n,whether it is effectively solved in JMP is a key exploration,and it is also to provide suitable predictive variables for genome-wide prediction in JMP,laying a solid foundation for predictive breeding.2.For 210 rice lines of rice materials,we used two common models in JMP,which are ridge regression with variable selection and Lasso with variable selection.These two methods have different effects on rice yield,thousand-grain weight,the number of tillers and the number of grains per plant can also be used for genome-wide prediction.The correlation coefficient between the predicted value and the actual value is between0.7 and 0.9.In this data material,we mainly explore the model comparison function in JMP.We use two generalized regression models,ridge regression and Lasso,to compare the genome-wide prediction results of rice yield and related traits in JMP.Function compare the differences between the prediction models and choose a more appropriate model.