vsab
New Member
Posts: 3
|
Post by vsab on Apr 9, 2015 10:47:31 GMT
Hi there
I am using GCTA to conduct mmla because it allows me to incorporate a relationship matrix as i have related individuals (not family data). I want to inlcude age, sex and PCA's in the analysis but i am struggling as to how to do it. Previous posts have suggested adjusting the phenotype for covariates (age and sex) prior to the running the flag mlma. What are the commands that i use and how should i construct the files, 2 separate because one is discrete and the other (age) is quantitative...
gcta64 --bfile [plinkfile] --make-grm --maf [e.g. 0.01] --out[filename]
gcta64 --grm [filename] --pca 10 --out [filename]
gcta64 --mlma --bfile [plinkfile] --grm [filename] --pheno [filename.phen] --mpheno [#] --qcovar [filename.eigenvec] --out
Where in the commands do i adjust the GRM for the other covariates i.e age and sex?
Please help. I am just starting out with GCTA
|
|
|
Post by Zhihong Zhu on Apr 9, 2015 22:57:19 GMT
Hello,
One way is to combine the age and sex with 10 PCs into a file, eg. covar.txt. When you are running MLMA, just typing "--qcovar covar.txt".
Cheers, Zhihong
|
|
vsab
New Member
Posts: 3
|
Post by vsab on Apr 11, 2015 4:49:37 GMT
Thanks, have tried that in the mean time. Is sex not a discreet variable and does GCTA read the 0/1 of Plink as a discreet variable or should one include it as you suggested in the qcovar file as quantitative?
Regards Venesa
|
|
|
Post by Zhihong Zhu on Apr 12, 2015 3:26:05 GMT
Hi Venesa,
Yes, gender is a discrete variable, please put that in another file, eg. "gender.covar", using the option, "--covar gender.covar".
Cheers, Zhihong
|
|
vsab
New Member
Posts: 3
|
Post by vsab on Apr 25, 2015 10:16:58 GMT
Hi There
Thanks alot. Is there someway of including confidence intervals when running mlma?
|
|
|
Post by Zhihong Zhu on Apr 26, 2015 11:59:39 GMT
Hi Venesa,
I'm afraid gcta doesn't provide the confident interval of beta for each SNP. The 95% confident interval is about [beta - se*1.96, beta+se*1.96], where beta is the estimate of effect size, and se is the SE of beta.
Cheers, Zhihong
|
|