Tweet
Hi all,
I am trying to do a GWAS in PLINK.
I am running
which give me an acceptable output with a subset of my desired covariants. However, I need to have both age and age^2 of the patients as covariants within the final analysis.
When I include both of these as covariants, PLINK provides 'NA' as the output for all association tests. This is a 'feature' of PLINK:
http://pngu.mgh.harvard.edu/~purcell...faq.shtml#faq6
The NA issue arises due to excessive correlation of two covariants, is my understanding, so as age and age^2 are inexorably linked, the --vif alteration cannot get around it.
I know that GWAS using these two covariants have been performed in PLINK successfully, based on some GWAS paper methods sections, but I cannot seem to work out a way to get this to work.
Does anyone have any suggestions as to how I may bypass this feature, I am sure that there are sound statistical logics as to why bypassing this may not be the soundest thing to do, but my analyses are going into a multi-centre meta-analysis, so I really don't get any say in what models are in use at this stage.
Any suggestions would be (extremely) gratefully received.
I am trying to do a GWAS in PLINK.
I am running
plink --file [file] --linear --covar [covariant-file] --out [name]
When I include both of these as covariants, PLINK provides 'NA' as the output for all association tests. This is a 'feature' of PLINK:
http://pngu.mgh.harvard.edu/~purcell...faq.shtml#faq6
The NA issue arises due to excessive correlation of two covariants, is my understanding, so as age and age^2 are inexorably linked, the --vif alteration cannot get around it.
I know that GWAS using these two covariants have been performed in PLINK successfully, based on some GWAS paper methods sections, but I cannot seem to work out a way to get this to work.
Does anyone have any suggestions as to how I may bypass this feature, I am sure that there are sound statistical logics as to why bypassing this may not be the soundest thing to do, but my analyses are going into a multi-centre meta-analysis, so I really don't get any say in what models are in use at this stage.
Any suggestions would be (extremely) gratefully received.