Abstract
Genomic prediction (GP) can help increase the efficiency of breeding programs, as genotypes can be selected based on their predicted performance. However, to the best of our knowledge, this procedure is not yet routine in commercial breeding programs in tetraploid organisms like potato (Solanum tuberosum L.). The objectives of this study were to (i) Estimate the prediction accuracy for 26 different potato traits in a panel of about 1000 genotypes based on 202,008 single nucleotide polymorphisms, (ii) Evaluate the influence of the size and constitution of the training set on the prediction accuracy, and (iii) Investigate how the effect of selection in the training set influences the outcome of GP. GP revealed high prediction accuracies using genomic best linear unbiased prediction. Our results indicated that a training set of 280-480 clones and 10,000 markers was sufficient. Prediction within a specific market segment led to a higher prediction accuracy compared to adding clones from other market segments to the training set or to predict between different market segments. Lastly, we found a higher prediction accuracy when in a training set of selected clones, i.e., a training set that consists of clones with high trait values, 20% of the clones were replaced by clones that were sampled from the clones that showed the lowest 10% trait values. This observation shows that clones from advanced breeding stages can be used as training set, if some clones specifically from the other side of the distribution range are added to the training set.