Improved Methods for Side Chain and Loop Predictions via the Protein Local Optimization Program:  Variable Dielectric Model for Implicitly Improving the Treatment of Polarization Effects

This paper presents significant improvements in both accuracy and computational efficiency of protein side chain and loop predictions using the Protein Local Optimization Program (PLOP). We introduce a novel energy model in which the internal dielectric constant of the protein is allowed to vary as a function of the interacting residues and present a physical rationale for this model. Using this model, we achieve qualitative improvements in the accuracy of side chain predictions with respect to experimental crystal structure and substantially reduce the RMSDs for loop predictions, particularly those predictions involving charged side chains. For the single side chain prediction of lysine, 40% of the errors are eliminated, and the accuracy increases from 62.6% to 76.8%. The errors in glutamate and aspartate predictions are reduced by 19% and 24%, respectively. When applied to a set of 240 loop predictions with 6, 8, 10, and 13 residue of loop length, this new model yields unprecedented accuracies with average backbone root-mean-square deviations of 0.39 Å, 0.68 Å, 0.80 Å, and 1.00 Å for 6, 8, 10, and 13 residue loops, respectively. We also describe a series of technical improvements in the PLOP simulation algorithms, which lead to a speedup of a factor of 2-4 in loop predictions.