This is a test version of Biostars. For the public version, visit https://www.biostars.org.
Dependent and independent covariates

Hi all,

For doing GWAS and eQTL analyses typically many covariates are included. Usually known covariates like gender, age etc. and also unknown covariates like PCA dimensions. Problem is some of these covariates might be not independent and have a relatively high correlation. Does anybody knows a what a common cut of value is for correlation or what to do with not independent covariates like age and PCA dimensions?

expression gene covariates
• 1,948 views • link
updated by Andrey Shabalin • Andrey Shabalin • written by Floris Brenk Floris Brenk

1 answer

I prefer to avoid using highly correlated covariates as they can often be redundant.

My approach is to use the following procedure:

  1. Select the (non-PCA) covariates that you are certainly including in the analysis.
  2. Perform PCA analysis on the data with the covariates selected in step 1 regressed out.
  3. Note that top PCs from step 2 are guarantied to be orthogonal to the covariates selected in step 1.
  4. Run eQTL analysis with covariates from step 1 and a few top PCs from step 2.
• 0 views • link
written by Andrey Shabalin • Andrey Shabalin

Log in to answer this question.

More posts like this