Data from: Leveraging historical trials to predict Fusarium head blight resistance in spring wheat breeding programs

Fusarium head blight (FHB) is a fungal disease posing a major threat to wheat production. Plant breeding that leverages genotyping is an effective method to improve the genetic resistance of cultivars. Started in 1995, the uniform regional scab nursery (URSN) consists of germplasm from several public breeding programs in the Northern U.S region. Its main objective is to showcase new sources of resistance and enable germplasm exchange among the cooperators, however, the data from the URSN has not been studied. Phenotypic and genotypic data from this nursery was gathered, as well as from two current breeding programs in the U.S Midwest. Genomic prediction on eight traits related to FHB and agronomic traits was applied, and the effects of statistical method, marker density, training set size, genetic structure, and genetic architecture of the trait were studied. Using the URSN population, RKHS was the best method in various prediction settings, with an average accuracy of 0.63, marker density could be as low as 500 without decreasing the prediction accuracy, and training set optimization was useful for two traits. Furthermore, genotypic values were predicted in breeding programs using the URSN population as a training set with various prediction scenarios. Predicting unrelated populations led to a significant decrease in accuracy but with encouraging values for some traits and populations. Ultimately, when progressively decreasing the number of lines from breeding populations in the training set, the advantage of adding the URSN population was more pronounced, with an increase in accuracy up to 0.19.小麦赤霉病（Fusarium head blight, FHB）是一类严重威胁小麦生产的真菌病害。借助基因分型（genotyping）技术开展的植物育种，是提升栽培品种遗传抗性的有效途径。统一区域赤霉病圃（uniform regional scab nursery, URSN）始建于1995年，其种质资源取自美国北部地区的多个公共育种项目，核心目标是展示新型抗源并推动合作方间的种质交流，但目前针对URSN产生的相关数据尚未得到系统研究。本研究收集了该病圃以及美国中西部当前两个育种项目的表型数据与基因型数据，针对与FHB相关的8个性状及农艺性状开展了基因组预测（genomic prediction）分析，探究了统计方法、标记密度、训练集规模、遗传结构以及性状遗传架构对预测效果的影响。基于URSN群体的预测结果显示，RKHS在多种预测场景下均为最优方法，平均预测精度达0.63；标记密度低至500时仍不会降低预测精度；且训练集优化可有效提升两个性状的预测性能。此外，本研究以URSN群体作为训练集，结合多种预测场景，对多个育种项目的育种材料开展了基因型值预测。对无关群体进行预测时，预测精度会出现显著下降，但部分性状与群体的预测结果仍令人满意。最终，当逐步缩减训练集内育种群体的品系数量时，加入URSN群体的优势会愈发显著，预测精度最高可提升0.19。