non-parameteric method

This study aimed to investigate the  effect  of  the method of estimating the effects of markers , QTLs distribution, number of QTLs, effective population size and trait heritability on the accuracy of genomic predictions. Two effective population sizes, 100 and 500 individuals, were simulated by QMSim software. A 100 cM genome including one chromosome was simulated where 500 SNPs and two different  numbers of QTLs (50 and 200) were distributed on it randomly. In this study three levels of heritability (0.1, 0.3 and 0.5) were considered. Genomic breeding values were predicted using Bayesian ridge regression, BayesA, BayesB, BayesC, Bayesian LASSO, Reproducing kernel Hilbert space and neural networks. In this research, the accuracy of genomic breeding values were affected by trait heritability, effective population size, markers effect estimation methods, QTLs distribution and number of QTLs. The Bayes A and B had the highest accuracy while accuracy of neural networks method was the lowest. The accuracy of genomic breeding values were increased as the heritability of trait and number of QTLs increased while the accuracy was decreased as the effective population decreased. Considering the QTLs distribution, the highest accuracy was achieved when the QTLs distributed normally.