参考群筛选方法及规模对基因型填充准确性的影响

为探究基于A矩阵期望遗传关系最大化（maximizing the expected genetic relationship for matrix A，RELA）、基于A矩阵目标群体遗传方差最小化（minimized the target population genetic variance for matrix A，MCA）、平均亲缘关系最大化（the highest mean kinship coefficients，KIN）、随机选择（random selection，RAN）、共同祖先筛选（common ancestor，CA）等不同参考群筛选方法及参考群规模对基因型填充准确性的影响。本研究使用矮小型黄羽肉鸡作为试验群体，采用鸡600K SNP芯片（Affymetrix Axion HD genotyping array）进行基因分型，测定435羽子代公鸡45、56、70、84、91日龄体重。利用Beagle软件将低密度SNP芯片填充为高密度SNP芯片数据，比较不同参考群筛选方法、参考群规模对基因型填充准确性的影响，以及填充芯片基因组预测准确性。结果表明，使用Beagle 4.0结合系谱信息进行填充效果最佳，其次为Beagle 4.0，而Beagle 5.1填充效果最差。使用MCA方法筛选参考群进行基因型填充准确性最高，使用RAN方法筛选参考群进行基因型填充准确性最低，MCA、RELA、CA 3种方法基因型填充准确性差别较小。相比其他方法，使用MCA方法筛选个体作为参考群将低密度SNP芯片填充至高密度SNP芯片进行基因组选择的预测准确性较高，与真实高密度SNP芯片的基因组预测准确性相差甚微。随着参考群规模增大，基因型填充准确性也随之增加，但增速逐渐下降，最后趋于平缓。综上所述，可以通过参考群筛选方法构建参考群以及控制参考群规模，以保证基因型填充和基因组预测准确性并节省成本，本研究为基因型填充在畜禽遗传育种中的应用提供技术参考。

Effect of Reference Population Selection Method and Size on Genotype Imputation Accuracy

The study aimed to explore the influence of different reference population screening methods and size on accuracy of genotype imputation, such as maximizing the expected genetic relationship for matrix A (RELA), minimized the target population genetic variance for matrix A(MCA), the highest mean kinship coefficients (KIN), random selection (RAN), common ancestor (CA). In this study, the dwarf and yellow-feathered chicken population were used, and the chicken 600K SNP array(Affymetrix Axion HD genotyping array) was used for genotyping. The body weight of 435 offspring cocks at 45, 56, 70, 84 and 91 days of age were measured. The Beagle software was used to impute low-density SNP chips into high-density SNP chips, to compare the influence of reference population screening methods and reference population size on accuracy of imputed genotype and accuracy of imputed chips for genomic prediction. The results showed that the best imputation method was using Beagle 4.0 with pedigree information, followed by Beagle 4.0, and Beagle 5.1 was comparatively the worst. MCA method had the highest accuracy of genotype imputation, RAN method had the lowest accuracy of genotype imputation, the accuracy of MCA, RELA and CA methods for genotype imputation had small difference. Compared with other methods, MCA method has higher prediction accuracy to select key individuals as reference population and to impute from low-density SNP chips to high-density SNP chips for genome selection, which was slightly different from that of real high-density SNP chips. With the increase of reference population size, the accuracy of genotype imputation were also increased, but the growth rate were gradually decreased and finally tended towards stability. In conclusion, the accuracy of genotype imputation and genome prediction,as well as lower costs were guaranteed by selecting key individual screening methods and controlling the size of reference population. This study provides technical reference for the application of genotype imputation in livestock genetic breeding.