Predicted values VS residuals

Dear all;

I am sorry for my earlier post without subject. My question in earlier mail
was:

I am going to run models for variable selection using elastic-net. Because
I have about 1500 descriptive (independent) variables which they are highly
correlated. Before running elastic-net and even single regressions (~1500)
I need get fitted values for dependent variables using MIXED model. My
question is:

Should I used the fitted values or residuals as dependent variable for run
elastic-net (even single regressions which is not main purpose)?

Your helps are highly appreciated,

Greg