# Adjusting phenotypes by regressing out covariates

I'm trying to use the bfGWAS tool, which analyses GWAS data and integrates functional annotations to identify casual SNPs (paper and github). In the user manual, it states:

We recommend first regressing covariants out from the original phenotypes, and then provide bfGWAS the corrected phenotypes (i.e., residuals from the regression model with covariates).

It is unclear to me what this means and how to go about determining the corrected phenotypes. Typically, in an association test with covariates, a regression would be performed using a model like

pheno ~ snp + covariate


I could then get the residuals from this model. However, I'm not certain how to go about it for all SNPs, and then adjust the phenotypes. I have case-control data, so the phenotypes are binary.

Instead of the "observed phenotype", it's asking for a "corrected phenotype", which is the residuals from pheno ~ covariates. In R, one would get that as follows:
fit = lm(phenotype ~ covariates, data)

Of course covariates is just a place hold for something like age + gender or something along those lines.