Toward correct protein folding potentials

Empirical protein folding potential functions should have a global minimum near the native conformation of globular proteins that fold stably, and they should give the correct free energy of folding. We demonstrate that otherwise very successful potentials fail to have even a local minimum anywhere near the native conformation, and a seemingly well validated method of estimating the thermodynamic stability of the native state is extremely sensitive to small perturbations in atomic coordinates. These are both indicative of fitting a great deal of irrelevant detail. Here we show how to devise a robust potential function that succeeds very well at both tasks, at least for a limited set of proteins, and this involves developing a novel representation of the denatured state. Predicted free energies of unfolding for 25 mutants of barnase are in close agreement with the experimental values, while for 17 mutants there are substantial discrepancies.