Heritability Calculation

Author

Natasha Santhanam

Published

June 3, 2021

How to calculate Heritability using GCTA

If you’re data is already in a PLINK form then it’s quite easy to use GCTA. You need to first create the genetic relatedness matrix (GRM) - make sure you’ve already done quality control on the genotype data since a high rate of missing SNPS will lead to negative eigenvalues in the GRM.

gcta –bfile ./plink_files –maf 0.01 –make-grm –out test –thread-num 10

The command above will produce the GRM genomewide. The genetic relationship matrix will be saved in the files test.grm.bin, test.grm.N.bin and test.grm.id in whatever directory you’re standing in. For datasets with an extremely large number of SNPs and large sample size (e.g. 1000G imputed data) you can use the flag –chr to calculate the GRM for each autosome. If you want to remove an relatedness, you can also use the flag –grm-cutoff (make sure what your cutoff is and that its not too strict)

For REML analysis, you next use the command below. GCTA uses the first column in the phenotype file as the phenotype it’s testing. If you want to use another column you can use the –mpheno flag.

gcta6 –reml –grm test –pheno test_cc.phen –out test_cc

Above is all for genomewide heritability. If you want to calculate cis-heritability, you need to make individual GRMs for each gene and only include the cis-snps. You would then run the above gcta command in a loop for each GRM. Here’s an example:

for(i in 1:length(genelist)){ cat(i,"/",length(genelist),"\n") gene <- genelist[i] geneinfo <- gtf[match(gene, gtf$gene_id),] chr <- geneinfo[1] c <- chr$chr start <- geneinfo$start - 1e6 ### 1Mb lower bound for cis-eQTLS end <- geneinfo$end + 1e6 ### 1Mb upper bound for cis-eQTLs chrsnps <- subset(bim,bim[,1]==c) ### pull snps on same chr cissnps <- subset(chrsnps,chrsnps[,4]>=start & chrsnps[,4]<=end) ### pull cis-SNP info snplist <- cissnps[,2] write.table(snplist, file= gt.dir %&% "tmp.cis.SNPlist",quote=F,col.names=F,row.names=F) runGCTAgrm <- "gcta --bfile " %&% box.dir %&% "tmp --make-grm-bin --extract " %&% gt.dir %&% "tmp.cis.SNPlist" %&% " --out " %&% grm.dir %&% gene system(runGCTAgrm) }

Another reminder to perform quality control on your genotype data before going forth in calculating heritability. For example, here is genotype data with a high rate of missing SNPS

0% 25% 50% 75% 100% ___ 0.07913669 0.09352518 0.10791367 0.12230216 0.71223022

Almost half of the samples contain only 10% of all snps. This genotype data would not be right to use for GCTA since the program will have issues calculating the GRM. It is better ot instead imput the genotype and then use that

--- title: Heritability Calculation author: Natasha Santhanam date: '2021-06-03' slug: heritability-calculation categories: [] tags: [] --- How to calculate Heritability using GCTA If you're data is already in a PLINK form then it's quite easy to use GCTA. You need to first create the genetic relatedness matrix (GRM) - make sure you've already done quality control on the genotype data since a high rate of missing SNPS will lead to negative eigenvalues in the GRM. gcta --bfile ./plink_files --maf 0.01 --make-grm --out test --thread-num 10 The command above will produce the GRM genomewide. The genetic relationship matrix will be saved in the files test.grm.bin, test.grm.N.bin and test.grm.id in whatever directory you're standing in. For datasets with an extremely large number of SNPs and large sample size (e.g. 1000G imputed data) you can use the flag --chr to calculate the GRM for each autosome. If you want to remove an relatedness, you can also use the flag --grm-cutoff (make sure what your cutoff is and that its not too strict) For REML analysis, you next use the command below. GCTA uses the first column in the phenotype file as the phenotype it's testing. If you want to use another column you can use the --mpheno flag. gcta6 --reml --grm test --pheno test_cc.phen --out test_cc Above is all for genomewide heritability. If you want to calculate cis-heritability, you need to make individual GRMs for each gene and only include the cis-snps. You would then run the above gcta command in a loop for each GRM. Here's an example: > `for(i in 1:length(genelist)){ cat(i,"/",length(genelist),"\n") gene <- genelist[i] geneinfo <- gtf[match(gene, gtf$gene_id),] chr <- geneinfo[1] c <- chr$chr start <- geneinfo$start - 1e6 ### 1Mb lower bound for cis-eQTLS end <- geneinfo$end + 1e6 ### 1Mb upper bound for cis-eQTLs chrsnps <- subset(bim,bim[,1]==c) ### pull snps on same chr cissnps <- subset(chrsnps,chrsnps[,4]>=start & chrsnps[,4]<=end) ### pull cis-SNP info snplist <- cissnps[,2] write.table(snplist, file= gt.dir %&% "tmp.cis.SNPlist",quote=F,col.names=F,row.names=F) runGCTAgrm <- "gcta --bfile " %&% box.dir %&% "tmp --make-grm-bin --extract " %&% gt.dir %&% "tmp.cis.SNPlist" %&% " --out " %&% grm.dir %&% gene system(runGCTAgrm) }` Another reminder to perform quality control on your genotype data before going forth in calculating heritability. For example, here is genotype data with a high rate of missing SNPS 0% 25% 50% 75% 100% ___ 0.07913669 0.09352518 0.10791367 0.12230216 0.71223022 Almost half of the samples contain only 10% of all snps. This genotype data would not be right to use for GCTA since the program will have issues calculating the GRM. It is better ot instead imput the genotype and then use that