Safflower, a minor oilseed crop, is gaining increased attention for food and industrial uses. Safflower genebank collections are an important genetic resource for crop enhancement and future breeding programs. In this study, we investigated the population structure of a safflower collection sourced from the Australian Grain Genebank and assessed the potential of genomic prediction (GP) to evaluate grain yield and related traits using single and multi‐site models. Prediction accuracies (PA) of genomic best linear unbiased prediction (GBLUP) from single site models ranged from 0.21 to 0.86 for all traits examined and were consistent with estimated genomic heritability (h²), which varied from low to moderate across traits. We generally observed a low level of genome × environment interactions (g × E). Multi‐site g × E GBLUP models only improved PA for accessions with at least some phenotypes in the training set. We observed that relaxing quality filtering parameters for genotype‐by‐sequencing (GBS), such as missing genotype call rate, did not affect PA but upwardly biased h² estimation. Our results indicate that GP is feasible in safflower evaluation and is potentially a cost‐effective tool to facilitate fast introgression of desired safflower trait variation from genebank germplasm into breeding lines.

中文翻译：

---

红花种质库中籽粒产量及其相关性状的基因组预测和基因组遗传力

红花是一种次要的油料作物，在食品和工业用途方面正受到越来越多的关注。红花种质库收藏是增强作物和未来育种计划的重要遗传资源。在这项研究中，我们调查了来自澳大利亚谷物基因库的红花集合的种群结构，并评估了基因组预测（GP）使用单点和多点模型评估谷物产量和相关性状的潜力。单点模型的基因组最佳线性无偏预测（GBLUP）的预测准确性（PA）范围从0.21到0.86，检查的所有性状均与估计的基因组遗传力（h ²），其特征从低到中等不等。我们通常观察到低水平的基因组×环境相互作用（g×E）。多站点g×E GBLUP模型仅针对在训练集中具有至少某些表型的种质改良了PA。我们观察到，针对基因型序列排序（GBS）的放宽质量过滤参数（例如缺失的基因型检出率）不会影响PA，但会影响h ²估计的偏向。我们的结果表明，GP在红花评估中是可行的，并且可能是一种经济高效的工具，可促进所需的红花性状变异从种质库种质快速进入种系。