Genetic Relation Matrix (GRM)

The GRAB package supports both dense GRM (for POLMM only) and sparse GRM to characterize family relatedness and help prevent inflated type I error rates. Dense GRM is calculated from genotype files when fitting the null model (POLMM). Sparse GRM is loaded from a whitespace-delimited file with three columns: ID1, ID2, and the genetic correlation between the two subjects.

Note: Based on simulation and real data analysis results, for binary and ordinal categorical data analysis, analyses using dense and sparse GRM perform similarly in terms of both type I error rates and statistical power.

How to make a Sparse GRM file

About function getSparseGRM

• GRAB package includes a function getSparseGRM, which implicitly uses GCTA software (we tested v1.93) to make a SparseGRMFile to be passed to function GRAB.NullModel.

• It has been reported that if the PLINK files include subjects with more than one ancestry, the GRM estimation might be highly inaccurate.

Step 0: Prepare PLINK binary files

PLINK binary files with high-quality genotyped variants are required to make a sparse GRM. In UK Biobank real data analysis, we used the following cutoffs in PLINK to select ~ 340K SNPs for White British subjects.

--maf 0.05
--indep-pairwise 500 50 0.2

Step 1: RUN getTempFilesFullGRM to get temporary files

• Besides PlinkPrefix, arguments nPartsGRM and partParallel are required.
• gcta64File sets the path to GCTA executable file, which is also required.
• The GRM calculation is split to nPartsGRM parts for parallel computation.
• For UK Biobank data analysis with ~ 500K samples, we recommend setting nPartsGRM = 250 and using multiple CPU cores in High Performance Cluster.
• If not specified, the temporary files are in tempdir(). Users can set tempDir to change it.
• If the sample size > 100K, then the temporary files might need a large amount of space.
• Other arguments includes
  • subjData: a character vector to specify subject IDs to retain. Default is NULL, i.e. all subjects are retained in sparse GRM.
  • minMafGRM: Minimal value of MAF cutoff to select markers (from PLINK files) to make sparse GRM. (default=0.01)
  • maxMissingGRM: Maximal value of missing rate to select markers (from PLINK files) to make sparse GRM. (default=0.1)
  • threadNum: Number of threads (CPU cores) to use.

Example:

library(GRAB)
GenoFile = system.file("extdata", "simuPLINK.bed", package = "GRAB")
PlinkPrefix = tools::file_path_sans_ext(GenoFile)   # remove file extension
nPartsGRM = 2;
for(partParallel in 1:nPartsGRM) {
  getTempFilesFullGRM(
    PlinkPrefix, 
    nPartsGRM = nPartsGRM, 
    partParallel = partParallel,
    gcta64File = "/path/to/gcta64"
  )
}

Step 2: RUN getSparseGRM to combine the temporary files

• Function getSparseGRM can search the temporary files from tempDir based on arguments minMafGRM and maxMissingGRM.
• Argument relatednessCutoff is the cutoff for sparse GRM, only kinship coefficient greater than this cutoff will be retained in sparse GRM. (default=0.05)
• If argument rm.tempFiles is set as TRUE, then all temporary files will be removed.

Example:

SparseGRMFile = system.file("extdata", "SparseGRM.txt", package = "GRAB")
getSparseGRM(PlinkPrefix, 
             nPartsGRM = nPartsGRM, 
             SparseGRMFile = SparseGRMFile,
             relatednessCutoff = 0.05)

About the SparseGRMFile

The below gives more details about the SparseGRMFile

SparseGRMFile = system.file("extdata", "SparseGRM.txt", package = "GRAB")
SparseGRM = data.table::fread(SparseGRMFile)
SparseGRM
#            ID1      ID2     Value
#    1:     f1_1     f1_1 0.9550625
#    2:     f1_2     f1_2 1.0272297
#    3:     f1_3     f1_3 1.0192574
#    4:     f1_4     f1_4 1.0053836
#    5:     f1_5     f1_1 0.4648096
#   ---
# 2547: Subj-496 Subj-496 1.0027448
# 2548: Subj-497 Subj-497 0.9913247
# 2549: Subj-498 Subj-498 0.9785985
# 2550: Subj-499 Subj-499 1.0109795
# 2551: Subj-500 Subj-500 0.9783296