Reliable protein structure refinement using a physical energy function

In the past decade, significant progress has been made in protein structure prediction. However, refining models to a level of resolution that is comparable with experimental results and can be used in studies like enzymatic activity still remains a major challenge. We have previously demonstrated that our modular protein-solvent energy function, uniquely involving a potential of mean force description for hydrophobic solvation, works well in protein globular structure prediction and loop modeling. In this work, we couple protein-solvent energy function with our global optimization method stochastic perturbation with soft constraints and use them to refine a collection of template models from submitted predictions to recent Critical Assessment of Techniques for Protein Structure Prediction blind prediction contests. A prediction protocol based on a selection of structures with the lowest energy is able to successfully refine all of the test proteins, and, more importantly, our energy function does not show degradation in prediction when sampling is exhausted.