As I understand it, once picked, fragments don’t have any sequence associated with them. They’re effectively disembodied backbone angles. (Which makes sense as the fragments don’t necessarily come from PDB proteins with exactly the input sequence – the input sequence in picking is mainly used for secondary structure prediction, and it’s that secondary structure that’s matched, rather than sequence.)
I also believe that in inserting fragments, the existing residues in the pose are used and the phi/psi angles are changed. (As opposed to creating new residues with the given backbone coordinates.)
So if you’re doing loop remodeling with something like a SER_phosphorylated, Rosetta shouldn’t have any problem with it, and should maintain the phosphorylated status of the residue.
Of course, you need to make sure your residue patch exists in both fullatom and centroid mode (as fragment manipulation typically happens in centroid mode), and that you’ve picked the fragments appropriately (I don’t think the fragment picker works with modified residues all that well, but the fragments picked with serine should be decent for phosphoserine, for example – the one trick would be if there’s different secondary structure propensities for the two, but in general you probably wouldn’t know that.)
.
