I have used backrub (Rosetta3.5) and it calculates scores for decoys.
When I recalculate scores using score.linuxgccrelease from the same rosetta version, different values (30% difference) are generated.
Both used score12 and and dun02. Why are the values so different?
Take a look at the individual scoreterm breakdown. Is there a particular term which is giving you the difference? It may be that the backrub application is adding additional scoreterms like constraints or chainbreak terms, and these may influence what your scores are like. Another possibility is that backrub may be reweighting a term internally. (In the output PDBs it should list what the associated scoring weights are.)
The other (remote) possibility is that the limited precision of the PDB has changed things. When the generating application calcuates things, it uses the full precision of the atom positions. But if you output the structure as PDBs and then rescore them, you're limited to 0.001 Ang precision. This rounding changes atom locations and thus energies. 30% difference is pretty massive for such a small movement of atoms, though, unless you're right on the edge of a major clash or something. I mention it as a possibility, but I really doubt that it's the case.
yes, although it claims SCOREFUNCTION to be score12, backrub indeed has an extra term mm_bend with weight of 1.0, which leads to the 1/3 difference.
This reweighting can be found also in tutorial http://rosettadesign.med.unc.edu/momeara/momeara_FeaturesAnalysisTutoria...
The mm_bend term is one which enforces reasonable geometry on the (three-center) bond angles. By default score12 doesn't have any enforcement of that geometry as typical usage of Rosetta doesn't perturb those degrees of freedom, manipulating torsional angles only. As backrub does slightly more extensive/different backbone geometry manipulation, turning on mm_bend to make sure that it doesn't mess up the backbone bond angles makes sense.