|
Post by porchard on May 25, 2016 1:18:36 GMT
Hello,
I have summary-level eQTL data and am wondering if cojo-slct would be appropriate to use to work out the SNPs independently associated with each gene. At the moment a median of 36 SNPs (covering a median distance of ~100kb) are significantly associated with each gene due to LD, and I'd like to narrow this down to a smaller region (I'm hoping to use these regions to assign enhancers to genes). So:
1) would this be an appropriate use of cojo-slct, or am I overlooking something,
2) the "effect size" is currently given in terms of the coefficient of LINEAR regression, as gene expression is not a binary trait -- is this acceptable, or will this break some assumption?
3) on the version of GCTA I have installed, it seems that --cojo-slct is no longer an acceptable option. I see in the source code that there is a "--massoc-slct" option -- is this interchangable with cojo-slct?
Thanks for your time,
porchard
|
|
|
Post by Jian Yang on May 25, 2016 12:50:37 GMT
Re 1) Yes, cojo-slct runs a stepwise model selection which would reduce the number of associated SNPs to one or a few if there are multiple signals.
Re 2) cojo-slct assume that the effect size is the coefficient from linear regression.
Re 3) They are the same. The source code in the website are a bit old. We will release the latest source code soon.
|
|