If you mean dG of folding, as in comparing an explicit folded state with an explicit unfolded state, then Rosetta sort of already did it, and sort of can’t do it.
1) The value reported by Rosetta’s standard scorefunction is implicitly a dG of folding. There is no unfolded state comparison, but the reference energies (ref) are meant to take the place of the unfolded state. You already have those numbers.
2) Generally, there is no way to do an explicit unfolded state in Rosetta – what would the state look like? There are an infinity of isoenergetic unfolded states to model.
3) There is an unfolded reference energy term that Doug Renfrew developed for use with noncanonical residue types, but it’s meant to replace the reference energies, so I don’t think it will do any better than the values you already have.
If you just want to compare the stabilities of your mutations to their parent sequences – run the parent sequence along with your mutations through the same protocol (you said you used backrub) and just subtract one from the other to get the ddG.