I'd like to include surrogate variables from SVA in a MatrixeQTL model to account for unknown sources of variation in the expression data.

SVA assumes that there are adjustment variables and variables of interest. The adjustment variables are the other covariates that I will include in the MatrixeQTL model (i.e. age and sex) but I'm unsure what to include as the variable of interest in this instance?

In this case there isn't a simple categorical variable of interest like "cancer status" in the SVA example.

I suppose the variables of interest in this case are the SNP dosages of the ~ 700,000 SNPs being included in the eQTL analysis. Are these the variables that should be included in the SVA full model? Or can surrogate variables be identified in SVA without a variable of interest?