Reply to: On powerful GWAS in admixed populations.
Reply to: On powerful GWAS in admixed populations.
复制标题
回复：关于混合人群中强大的 GWAS。
DOI：
10.1038/s41588-021-00975-z
复制
发表时间：
2021
期刊：
影响因子：
30.8
通讯作者：
Neale,BenjaminM
中科院分区：
文献类型：
--
作者：
Atkinson,ElizabethG;Bloemendal,Alex;Maihofer,AdamX;Nievergelt,CarolineM;Daly,MarkJ;Neale,BenjaminM
混合人群在基因组学研究中的代表性严重不足 1, 2。为了确保医学遗传突破公平地惠及所有血统的个体 3，需要开发有助于研究多样化和混合人群的工具。我们的文章 4 提出了一种通过纳入当地血统将混合个体纳入经过良好校准的全基因组关联研究 (GWAS) 的方法。在他们的评论中，侯等人。认为不包括当地血统的替代 GWAS 方法可以在不同血统的效应大小相等的情况下获得更高的功效。我们希望澄清的是，虽然我们确实观察到由于这种边缘情况下估计的参数数量增加而导致功率下降（图 2 中的参考文献 4、扩展数据图 3 和图 4 以及讨论），但在所有其他建模场景中，我们观察到功率提升。鉴于次要等位基因频率和连锁不平衡模式在全基因组人群之间通常存在差异 5,6,7,8，并且病例确定、上位性和基因-环境相互作用的差异可能因祖先而异，并导致边际效应大小差异，我们预计，即使假设因果效应相同，边际效应大小完全相同的情况也是例外，而不是规则。因此，预计 Tractor 在大多数 GWAS 位点上都优于其他方法。“边际效应”是指 GWAS 式单变量回归（包括分层控制）的估计量（大样本限制）。通过“因果效应”，我们指的是等位基因替代对等基因（和等环境）背景的影响。因果效应可能无法从观察到的数据中识别出来；如果所有变异都经过测量和索引，没有完美的连锁不平衡（LD），并且没有群体分层，那么它们将代表完整多元回归的估计值或大样本极限。边际效应向量通过乘以 LD 矩阵与因果效应向量相关。
Admixed populations are sorely underrepresented in genomics research 1, 2. To ensure that medical genetic breakthroughs equitably benefit individuals of all ancestries 3, there is a need for the development of tools that facilitate the study of diverse and admixed populations. Our article 4 proposes a methodology for the inclusion of admixed individuals in well-calibrated genome-wide association studies (GWAS) through the incorporation of local ancestry. In their comment, Hou et al. argue that alternative GWAS methods that do not include local ancestry can attain improved power in circumstances in which the effect sizes are equivalent across ancestries. We wish to clarify that while we indeed observe a power drop due to the increase in the number of parameters estimated in this edge case (addressed in ref. 4 in Fig. 2, Extended Data Figs. 3 and 4, and Discussion), in all other scenarios modeled we observe a power boost. Given that minor allele frequencies and patterns of linkage disequilibrium regularly differ between populations genome-wide 5, 6, 7, 8, and that differences in case ascertainment, epistasis and gene–environment interactions may differ across ancestries and induce marginal effect size differences, we expect the instance of perfectly identical marginal effect sizes to be the exception, not the rule, even assuming identical causal effects. Tractor is therefore expected to outperform other methods at most GWAS loci.By ‘marginal effect’, we mean the estimand (large-sample limit) of GWAS-style single variant regression (including control for stratification). By ‘causal effect’, we mean the effect of allelic substitution on an isogenic (and iso-environmental) background. The causal effects may be unidentifiable from observed data; if all variation were measured and indexed with no perfect linkage disequilibrium (LD), and there were no population stratification, then they would represent the estimand or large-sample limit of the full multivariate regression. The vector of marginal effects is related to the vector of causal effects by multiplication by the LD matrix.