时间戳

首页 > 解决方案 > MCMCglmm with genomic relatedness matrix: "levels do not have a row entry in ginverse"

r - MCMCglmm with genomic relatedness matrix: "levels do not have a row entry in ginverse"

问题描述

I am trying to estimate heritability of a quantitative trait using the animal model. Because I am studying a wild species of rodent, I lack pedigree information; instead, I am using a genomic relatedness matrix (GRM) using ~23,000 SNPs derived from ddRAD and estimated with PLINK/GCTA.

Although I have cross-checked the individuals in my GRM and my data file, I receive this error: 

Error in MCMCglmm(trait ~ 1, random = ~ID, data = phens, ginverse = list(ID = gctaGRM),  : 
  some levels of ID do not have a row entry in ginverse

This problem is similar to one described previously: MCMCglmm data format difficulties. In this case, the poster used a pedigree.

My data are as follows:

  1. Data file with individual id, site, and phenotypic data, "phens"
> class(phens)
[1] "data.frame"

> head(phens)
    Site      ID       trait
1   QERC SGTm037  0.94770905
2 Robles SGTm038 -0.53410457
3   QERC SGTm039  0.35680224
4   QERC SGTm040 -0.24319776
5     MH SGTm041  0.02952266
6     MH SGTm043  2.25680224
>
  1. Trait as a variable
> trait <- phens$trait # quantitative trait

> head(trait)
[1]  0.94770905 -0.53410457  0.35680224 -0.24319776  0.02952266  2.25680224
  1. GRM from either PLINK or GCTA, read in using gap package: https://github.com/cran/gap/blob/master/R/MCMCgrm.R
gctaGRM <- ReadGRM('gcta-GRM') # reading in GCTA grm
plinkGRM <- ReadGRMPLINK('plinkgenome') # reading in PLINK PIHAT grm

> head(gctaGRM)

          SGTm037     SGTm038     SGTm039
SGTm037 1.0487960 0.000000000 0.000000000
SGTm038 0.0000000 1.065967000 0.009045295
SGTm039 0.0000000 0.009045295 1.032933000
SGTm040 0.1843918 0.000000000 0.019915070
SGTm041 0.0000000 0.031226350 0.000000000
SGTm043 0.0000000 0.038495810 0.000000000

> head(plinkGRM)

        SGTm037 SGTm038 SGTm039 SGTm040 SGTm041
SGTm037  1.0000       0       0  0.1345       0
SGTm038  0.0000       1       0  0.0000       0
SGTm039  0.0000       0       1  0.0000       0
SGTm040  0.1345       0       0  1.0000       0
SGTm041  0.0000       0       0  0.0000       1
SGTm043  0.0000       0       0  0.0000       0

As far as I can tell, the list of IDs in phens agrees with the columns and rows of IDs in either GRM. I can provide the full data files for those who are curious. 

> head(phens$ID)
[1] SGTm037 SGTm038 SGTm039 SGTm040 SGTm041 SGTm043

# also played with changing phens$ID from factor to character; made no difference

colnames(gctaGRM)
[1] "SGTm037" "SGTm038" "SGTm039" "SGTm040" "SGTm041" "SGTm043"

This must be a formatting issue, and I spent a lot of time researching this on stack, google groups, and r documentation to no avail. Unfortunately, few resources cover how to implement GRMs in these models (though see https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0197720). I tried changing classes of my data file and trait, but this did not work.

Any suggestions would be greatly appreciated!

标签： rmodelingglmmcmc

解决方案

我收到了同样的错误，和你一样，我很困惑，因为 GRM 的行名确实与我的数据中的内容匹配。

就我而言，我通过使用标准数据框而不是 tidyverse tibble 来解决这个data问题。当 tibble 用于 时，MCMCglmm 不会抛出错误data，但不幸的是，其内部数据检查的编码方式不适用于 tibble。

推荐阅读