|
Post by brittany on Mar 25, 2014 20:44:22 GMT
Hello,
This is probably a naive question, but I just want to make sure that I am running my analysis correctly. I want to include both covariates (ex: gender) and quantitative covariates that include PCs to control for ancestry and other quantitative covariates like age. For the analysis I will create 2 documents: one with all the covariates (data.covar) and one with all the quantitative covariates (data1.qcovar) The quantitative covariates would then include all 10 PCs that I wish to control for in regards to ancestry and other quantitative covariates like age. For each of files they will also include the ids. Is that the correct way to make the files to control for covariates?
Thank you!
Brittany
|
|
|
Post by Jian Yang on Mar 26, 2014 10:39:27 GMT
Yes, that's right. However, I would adjust phenotype for age and sex prior to the REML analysis, and include only the PCs as covariates in the REML analysis.
|
|
|
Post by brittany on Mar 26, 2014 14:20:10 GMT
Thank you for the quick reply! So would you suggest only the PCs as covariates in the analysis. Why is that?
|
|
|
Post by Zhihong Zhu on Mar 27, 2014 3:11:58 GMT
It is divided into two steps, 1. "adjust phenotype for age and sex prior to the REML analysis" 2. "include only the PCs as covariates in the REML" Because phenotypes in the 2nd step has been normalised, i.e. the mean ( as well as variance ) of phenotypes in male and female group are the same.
|
|
|
Post by brittany on Mar 27, 2014 15:23:41 GMT
Thanks!
|
|
|
Post by Jaden on May 20, 2014 20:42:51 GMT
If I have more covariates besides age and gender I need to adjust, should I also adjust all of them prior to the REML analysis? Thank you
|
|
|
Post by Jian Yang on May 22, 2014 0:14:05 GMT
You don't have to. I usually adjust the phenotype for age and gender prior to the REML analysis and fit some other covariates in the REML analysis.
|
|
|
Post by Jaden on Jun 10, 2014 21:47:21 GMT
Hi, if my outcome is binary, for running the REML analysis, do I still need to adjust age and gender first? There is no residual for binary outcome after adjusting outcome. Thanks.
|
|
|
Post by Jian Yang on Jun 18, 2014 2:52:02 GMT
You can actually include the covariates in the GREML analysis but you need to make sure that there is no co-linearity problem of the covariates.
|
|
|
Post by New_Asling on Jun 18, 2014 9:20:29 GMT
Can I ask which method you use to adjust phenotype for age and sex ? Thank you!
|
|
|
Post by Jian Yang on Jun 18, 2014 11:20:16 GMT
Regressing phenotype on age and standardising the residuals in each gender group.
|
|
|
Post by Jaden on Jun 18, 2014 22:56:13 GMT
Thanks for your response, Jian. Just to be clear, is GREML analysis you mentioned is new command in the newest version of the software or is it a new name for the original command -reml? Thank you.
|
|
|
Post by Jian Yang on Jun 19, 2014 0:21:05 GMT
GREML is the name we came up recently for the method. Previously people call the GRM + REML approach the GCTA method. However, GCTA has now been able to do many different things. We have now started to name each specific analysis/method that you can do in GCTA.
|
|
|
Post by Jaden on Jun 24, 2014 21:09:18 GMT
Hi, I have a follow-up question related to GREML for binary outcome in GCTA. Should the sum of the proportion explained by genetic variance on each chromosome be the same as the proportion explained by genetic variance using one big file with chr1-22? For example, I ran GREML for each chromosome separately, I summed up the proportion and got about 22% explained and I also ran GREML using one file with all 22 chromosomes, I got about 12%. Assuming each chromosome is independent and using same number of covariates, I would expect the results from those two runs will be very similar. why they are so different?Thanks.
|
|
|
Post by Jian Yang on Jun 25, 2014 6:48:30 GMT
Please see Yang et al. 2011 Nat Genet. www.nature.com/ng/journal/v43/n6/abs/ng.823.htmlChromosomes are independent if there is no population structure but there will be a between-chromosome correlation at the presence of cryptic relatedness and/or population stratification. You can actually do a joint analysis of all 22 chromosomes using the --mgrm option in the REML analysis.
|
|