Use of different marker pre-selection methods based on single SNP regression in the estimation of Genomic-EBVs

Two methods of SNPs pre-selection based on single marker regression for the estimation
of genomic breeding values (G-EBVs) were compared using simulated data provided by the
XII QTL-MAS workshop: i) Bonferroni correction of the significance threshold and ii) Permutation test
to obtain the reference distribution of the null hypothesis and identify significant markers at P<0.01
and P<0.001 significance thresholds. From the set of markers significant at P<0.001, random subsets
of 50% and 25% markers were extracted, to evaluate the effect of further reducing the number of
significant SNPs on G-EBV predictions. The Bonferroni correction method allowed the identification
of 595 significant SNPs that gave the best G-EBV accuracies in prediction generations (82.80%). The
permutation methods gave slightly lower G-EBV accuracies even if a larger number of SNPs resulted
significant (2,053 and 1,352 for 0.01 and 0.001 significance thresholds, respectively). Interestingly,
halving or dividing by four the number of SNPs significant at P<0.001 resulted in an only slightly decrease
of G-EBV accuracies. The genetic structure of the simulated population with few QTL carrying
large effects, might have favoured the Bonferroni method.