High cane yield and commercially extractable sucrose (CCS) content are two of the key sugarcane commercial traits selected in sugarcane breeding programs. Advancements in genomic prediction may provide opportunities to speed up gains for these traits in breeding programs by combining accurate prediction of breeding values in candidate parent clones shortening generation intervals. Selection trials in commercial breeding programs may provide training populations for developing genomic predictions. In this study, three different populations of clones in early and advanced stage selection trials in an established commercial sugarcane breeding program were used to assess genomic prediction accuracy. The clones (genotypes) were evaluated for cane yield and sugar content in field trials and genotyped using a SNP array developed for sugarcane cultivars and parents. Five models (Bayes A, Bayes B, Bayesian LASSO, Bayesian GBLUP and RKHS) were tested using pedigree and/or marker data. Prediction models that included marker information had higher prediction accuracies than models with pedigree data only. For CCS, the prediction accuracies for genotypes in advanced stage trials using DNA markers were superior compared with prediction accuracies for early-stage trials, suggesting that prior intensive selection for CCS did not diminish accuracy of genomic prediction. However, by contrast, for cane yield, the prediction accuracies were much less for the population in the advanced stages of selection. The levels of prediction accuracy obtained in most datasets (0.25–0.45) are encouraging for developing applications of genomic prediction to predict breeding values of yield and sugar content in sugarcane breeding programs.

高甘蔗产量和可商业提取的蔗糖（CCS）含量是在甘蔗育种计划中选择的两个关键甘蔗商业性状。通过结合候选亲本克隆中育种值的准确预测，缩短了世代间隔，基因组预测方面的进展可能为加速育种程序中这些性状的获得提供了机会。商业育种计划中的选择试验可能会为制定基因组预测提供培训人群。在这项研究中，在已建立的商业甘蔗育种计划的早期和晚期选择试验中，使用了三个不同的克隆种群来评估基因组预测的准确性。在田间试验中评估克隆（基因型）的甘蔗产量和糖含量，并使用为甘蔗品种和亲本开发的SNP阵列进行基因分型。使用谱系和/或标记数据测试了五个模型（贝叶斯A，贝叶斯B，贝叶斯LASSO，贝叶斯GBLUP和RKHS）。包含标记信息的预测模型比仅具有谱系数据的模型具有更高的预测准确性。对于CCS，使用DNA标记的晚期试验中基因型的预测准确性要优于早期试验中的基因预测准确性，这表明对CCS的先前密集选择不会降低基因组预测的准确性。但是，相比之下，对于甘蔗产量，处于选择后期的人群的预测准确性要低得多。