I have designed some protein binders based on pre-structured models.
In Rosetta, ddg for designed protein binders are about -20 r.e.u. The problem is that when I assessed them with some servers like PRODIGY (http://milou.science.uu.nl/services/PRODIGY/) predicting the binding affinity in protein-protein complexes, unbelievably the ΔG (kcal mol-1) for designed proteins are competently more than pre-structured models. For example, ΔG for designed protein and pre-structured model are -7.6 kcal mol-1 and -12.6 kcal mol-1 respectively. As far as I know, the more negative Gibbs free energy (ΔG) results in the formation of stronger complexes.
I do not understand what is the problem.