Question about interpreting parameters from linear models

[Hartecky]

Hello,

thank you for your work, much appreciate such comparison and evaluation of the methods for cell type deconvolution.

However I have a question about the final dataframe preparation and interpreting the results:

For example for ridge, lasso, elastic_net

RESULTS = apply(RESULTS,2,function(x) ifelse(x < 0, 0, x)) #explicit non-negativity constraint
RESULTS = apply(RESULTS,2,function(x) x/sum(x)) #explicit STO constraint

I don't understand why negative coefficients are zeroed? Isn't negative coefficient telling us that certain cell types are fewer compared to the reference, rather than thinking that certain cell types are absent? Help me understand this procedure

Thanks in advance!