Performance evaluation of support vector machine (SVM)-based predictors in genomic selection

Abstract

The aim was to compare predictive performance of SVM-based predictors constructed using different kernel functions (radial, sigmoid, linear and polynomial) in different genetic architectures of a trait (number of QTL, distribution of QTL effects) and heritability levels. To this end, a genome comprised of five chromosomes, oneMorgan each, was simulated on which 10,000 bi-allelic single nucleotide polymorphisms (SNP) were distributed.Cross validation employing a grid search was used to tune the meta-parameters of each kernel function. Pearson’scorrelation between the true and predicted genomic breeding values (rp,t) and mean squared error of predictedgenomic breeding values (MSEp) were used, respectively, as measures of the predictive accuracy and the overallfit. Meta-parameter optimization had a significant effect on predictive performance of SVM-based predictors insuch a way that by using improper meta-parameters, the predictive power of models decreased significantly. In allmodels, the accuracy of prediction increased following increase in heritability and decrease in the number ofQTLs. In most of scenarios, radial- and sigmoid-based SVM predictors outperformed polynomial and linear models.The linear-and polynomial-based SVM had lower rp,t and higher MSEp and, therefore, were not recommended forgenomic selection. The prediction accuracy of radial and sigmoid models was approximately the same in most ofthe studied scenarios; however, considering all pros and cons of radial and sigmoid kernels, radial kernel wasrecommended as the best kernel function for constructing SVM. All of studied SVM-based predictors were efficientusers of time and memory.