Dig into the ab initio application and see which scorefunction it’s using.
The low-numbered scorefunctions available as weight sets (score0, score1, up through maybe score5?) are ported from Rosetta++’s ab initio. You can see their contents at rosetta_database/scoring/weights. I assume they’re also used in Rosetta3’s ab initio.
Generally centroid and fullatom scorefunctions have different terms – compare score4 (centroid) with score12 (fullatom standard) to see what I mean.
If you have a mostly unstructured protein, Rosetta is not going to be able to do much with it. There is a “REPLONLY” score function hack from Ray Wang in the Baker lab meant to allow unstructured regions to be repulsive-only. I have no idea if it’s released or works in pyrosetta.
.
