GWAS分析中SNP解释百分比PVE | 1,SNP解释百分比之和为何大于1? #2021.12.21
关于GWAS分析中PVE的计算方法:
我查了一下,大体计算PVE的方法有三种:第一种回归分析或者方差分析的方法,计算R方(GLM模型),第二种是根据effect,se,maf计算PVE,第三种是根据LMM的矩阵构建计算PVE。
汇总如下:
所以准备研究一下。
先看一个常见问题:GWAS分析中,SNP解释百分比(PVE)之和为何大于1?
问题来源:TASSEL的google group:
https://groups.google.com/g/tassel/c/v1aOPCYFyJE/m/HromqNnNIyMJ
问题描述:
Dear all, Again I have a question. In association mapping results in Tassel by MLM model gives Marker R2 values. If Marker R2 reports proportion of phenotypic variation (between 0 to 1) that is explained by corresponding maker, why sum of marker R2 values for all of markers is more than 1?
为何GLM或者MLM中的Marker R2之和会大于1,如果R2是解释的百分比,那应该是在0~1之间呀。
张志武老师回复:
The association studies in TASSEL (MLM or GLM) are performed on markers one at a time. Therefore, the sum of the R square (R2) of markers could be bigger than 100%. One of the reasons is due to linkage disequilibrium (LD) between markers. For example, if a marker has R2 of 20% and the marker is in complete LD with other five markers, then the five markers will have R2 sum to 120%.
在TASSEL中GLM或者MLM模型中,是单标记扫描,之所以SNP的R2(R square)之和会大于1,因为标记间存在LD,比如一个标记关联的基因能解释20%的变异,这个位点附近有6个标记都存在LD状态,那么这6个标记的解释百分比之和就会是120%。
下一节介绍一下GAPIT中GLM的PVE的计算方法,并用R语言实现。