Assessing the Performance of Ridge Regression Method-6 in Genomic Evaluation of Discrete Threshold Traits with Additive and Dominance Genetic Architecture

Extended abstract Introduction and Objective: One of the important issues in genomic selection is estimating the effect of markers. In recent years, various methods have been proposed to estimate the effect of markers, each of which estimates genomic breeding values with different accuracy. One of the methods used in genomic evaluation is the ridge regression (rrBLUP), which has been used in different studies to predict genomic breeding values. Recently, by applying changes in the parameters of the rrBLUP method, a variety of this method so called Ridge Regression method 6 (RR-m6) has been proposed to solve regression problems. However, so far, this method has not been used in the genomic evaluation of threshold traits with additive and dominant genetic architecture, and its performance in this field is not known. Therefore, in this research, the prediction performance of this method was compared with other common methods of genomic evaluation. Material and Methods: A genome consisting of 10 chromosomes, each containing 1000 bi-allelic single nucleotide polymorphism (SNP) was simulated at heritability level of 0.5. All quantitative trait loci (QTLs) were given additive genetic effects, and their effects were modeled by gamma distribution. Two scenarios of the number of QTL were considered as 1 and 10% of the total number of SNPs (100 and 1000 QTL, respectively). Also, in different scenarios, 0.0, 50 and 100% of QTLs were given dominance effect. Genomic breeding values were estimated using RR-m6, rrBLUP, GBLUP, BayesA, regression tree (RT), Random Forest (RF) and boosting, and the indicators of LR method, including prediction accuracy, bias and dispersion or inflation of genomic breeding values (inflation) were used to analyze the breeding values estimated by different methods. In addition, the computing time and the amount of memory needed to execute the codes of each method on the CPU were calculated. Results: The results showed that the use of a purely additive model when the genetic dominance effects contributed to the phenotypic variation of the trait lead to decrease in accuracy and increase in the bias and dispersion of the genomic breeding values, amount of which depend on the number of QTLs that have dominance effect. Compared to other methods, the RR-m6 showed a very good performance, so that in all the studied scenarios, estimated genomic breeding values by RR-m6 had the highest accuracy and the lowest bias and dispersion, although in most cases the differences were not significant with BayesA. In terms of computational speed, the RR-m6 method was the fastest, and compared to other methods, it required less memory to perform the analysis. Conclusion: The results showed that since the RR-m6 method predicted genomic breeding values with a high accuracy, and in the mean time it was very efficient in terms of the computing time and the required memory, it can be used for genomic evaluation of threshold traits.