GWAS计算PVE

GWAS结果如何计算SNP解释表型方差的比例(the proportion of phenotypic variance explained ,PVE )。理论上所有位点的 PVE 之和不能超过遗传力,但是实际计算得到的 PVE 之和会远大于1,因为位点之间存在连锁,而一般计算PVE都是每个位点单独计算的。

这里我整理了4种方法,安装使用的常用顺序排序如下

第一种

这种方法需要用到SNP效应值的 se 值,也是混合线性模型使用最广的方法。

基于文献1,PVE计算公式如下图中的红框

其中

  • 是标记效应值
  • 是标记效应值的标准误
  • 是最小等位基因型
  • 为样本数目

第二种

基于文献2,这个公式和上面的分子相同,只是分母直接用的表型方差,而不是上面的展开式子。

第三种

这种不是估计单个位点的PVE ,而是一堆位点的 PVE

在HIBLUP软件的答疑部分,建议是用显著位点和其它位点构建2个G阵,然后计算方差组分,最后基于方差组分结果计算这个

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
#step1: construct GRM1 using the significant SNPs
./hiblup --make-xrm
--bfile demo
--extract snp.txt
--out grm1
--threads 32 # grm1.GA.id, grm1.GA.bin will be generated
#step2: construct GRM2 using the SNPs excluding the significant SNPs
./hiblup --make-xrm
--bfile demo
--exclude snp.txt
--out grm2
--threads 32 # grm2.GA.id, grm2.GA.bin will be generated
#step3: compute the variance components for GRM1 and GRM2 by single trait model
./hiblup --single-trait
--pheno demo.phe
--pheno-pos X
--xrm grm1.GA,grm2.GA
--out vc
--threads 32
#step4: calculate the PVE for significant SNPs by following equation:
pve = V(grm1) / ( V(grm1)+V(grm2)+V(e) )

第四种

对于一般线性模型,可以利用一般线性回归的 R2 (决定系数)作为 PVE 结果。

具体计算的时候,也可以计算单独加上SNP的 R2 增量作为 PVE 结果,步骤如下

  1. 计算 SNP 和 其它固定因子在一般线性回归模型(R中的lm函数)对表型的决定系数 R2_all
  2. 剔除SNP,计算其它固定因子在一般线性回归模型(R中的lm函数)对表型的决定系数 R2_less
  3. PVE = R2_all - R2_less

参考文献

  1. Shim H, Chasman D I, Smith J D, et al. A multivariate genome-wide association analysis of 10 LDL subfractions, and their response to statin treatment, in 1868 Caucasians[J]. PloS one, 2015, 10(4): e0120758.
  2. Zhang T, Gao H, Sahana G, et al. Genome‐wide association studies revealed candidate genes for tail fat deposition and body size in the Hulun Buir sheep[J]. Journal of Animal Breeding and Genetics, 2019, 136(5): 362-370.
  3. https://www.hiblup.com/tutorials
  • 版权声明: 本博客所有文章除特别声明外,著作权归作者所有。转载请注明出处!
  • Copyrights © 2019-2026 Vincere Zhou
  • 访问人数: | 浏览次数:

请我喝杯茶吧~

支付宝
微信