ab99
New Member
Posts: 8
|
Post by ab99 on Mar 20, 2015 15:59:27 GMT
Hi,
We are trying to do quantitative trait prediction in an independent cohort using GCTA. We followed GCTA's online manual and did the following:
a) Use "--blup-snp" to calculate the BLUP solutions for the SNP effects; b) Predict the total genetic effect of individuals in an independent sample by PLINK --score option (i.e., add up all SNPs' effects together)
Two quick questions:
1) I believe the "total genetic effect of individuals" as estimated by "PLINK --score" does not include the effects of covariates. Am I right? 2) If I am right with 1), then the question is whether there is a way to add the effects of covariates into the predictor. Ultimately, we want to get the best prediction that we can, and often the covariates can contribute significantly.
Thank you! ab99
|
|
|
Post by Zhihong Zhu on Mar 21, 2015 3:42:38 GMT
|
|
ab99
New Member
Posts: 8
|
Post by ab99 on Mar 24, 2015 2:09:09 GMT
Hi Zhihong, Thank you for your reply! I understand that the "--reml-pred-rand" will provide the total genetic effect of each individual, but it provides the information for individuals that are used in the model fitting, plus it can not include the effects of covariates. So it can not achieve my goal. I also want to ask for a favor: could you confirm that per SNP BLUP in GCTA ("--blup-snp" option) considers the SNP effects as random effects? Thank you very much! ab99
|
|
|
Post by Zhihong Zhu on Mar 25, 2015 2:55:39 GMT
Hi ab99,
I'm sorry I didn't read your questions carefully.
1.Yes, in GCTA the SNP effects are considered as random effects. 2. I'm not sure how to include the effects of covariates in a simple way. But a way is conduct X*b by R, X - coefficients of the covariates, b - effects of covariates estimated by GCTA "--reml-est-fix".
Cheers, Zhihong
|
|
ab99
New Member
Posts: 8
|
Post by ab99 on Mar 25, 2015 18:28:43 GMT
Hi ab99, I'm sorry I didn't read your questions carefully. 1.Yes, in GCTA the SNP effects are considered as random effects. 2. I'm not sure how to include the effects of covariates in a simple way. But a way is conduct X*b by R, X - coefficients of the covariates, b - effects of covariates estimated by GCTA "--reml-est-fix". Cheers, Zhihong Hi Zhihong, Thanks a lot for the reply! Your proposal can solve our problem. I will go ahead and try the "--reml-est-fix" option. Thanks again, ab99
|
|