Optimizing genomic prediction model given causal genes in a dairy cattle population

文献类型： 外文期刊

第一作者： Teng, Jinyan

作者： Teng, Jinyan;Huang, Shuwen;Chen, Zitao;Ye, Shaopan;Diao, Shuqi;Yuan, Xiaolong;Zhang, Hao;Li, Jiaqi;Zhang, Zhe;Gao, Ning;Ding, Xiangdong

作者机构：

关键词： genomic selection; whole-genome sequence data; causal gene; prior knowledge

期刊名称：JOURNAL OF DAIRY SCIENCE （ 影响因子：4.034； 五年影响因子：4.354 ）

ISSN： 0022-0302

年卷期： 2020 年 103 卷 11 期

页码：

收录情况： SCI

摘要： As genotypic data are moving from SNP chip toward whole-genome sequence, the accuracy of genomic prediction (GP) exhibits a marginal gain, although all genetic variation, including causal genes, are contained in whole-genome sequence data. Meanwhile, genetic analyses on complex traits, such as genome-wide association studies, have identified an increasing number of genomic regions, including potential causal genes, which would be reliable prior knowledge for GP. Many studies have tried to improve the performance of GP by modifying the prediction model to incorporate prior knowledge. Although several plausible results have been obtained from model modification or strategy optimization, most of them were validated in a specific empirical population with a limited variety of genetic architecture for complex traits. An alternative approach is to use simulated genetic architecture with known causal genes (e.g., simulated causative SNP) to evaluate different GP models with given causal genes. Our objectives were to (1) evaluate the performance of OP under a variety of genetic architectures with a subset of known causal genes and (2) compare different GP models modified by highlighting causal genes and different strategies to weight causal genes. In this study, we simulated pseudo-phenotypes under a variety of genetic architectures based on the real genotypes and phenotypes of a dairy cattle population. Besides classical genomic best linear unbiased prediction, we evaluated 3 modified OP models that highlight causal genes as follows: (1) by treating them as fixed effects, (2) by treating them as a separate random component, and (3) by combining them into the genomic relationship matrix as random effects. Our results showed that highlighting the known causal genes, which explained a considerable proportion of genetic variance in the GP models, increased the predictive accuracy. Combining all given causal genes into the genomic relationship matrix was the optimal strategy under all the scenarios validated, and treating causal genes as a separate random component is also recommended, when more than 20% of genetic variance was explained by known causal genes. Moreover, assigning differential weights to each causal gene further improved the predictive accuracy.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序