Conformational Sampling in Structure Prediction and Refinement with Atomistic and Coarse-Grained Models

The performance of models at different resolutions is compared in the context of an iterative protein structure refinement protocol. The models considered here consist of an all-atom model described with the CHARMM22 force field in combination with a distance-dependent dielectric implicit solvent approximation, a united-atom model described by the CHARMM19 force field in combination with the effective energy function 1 (EEF1) solvent model, the new intermediate coarse-grained model PRotein Intermediate MOdel (PRIMO), both with Generalized Born and distance-dependent dielectric solvent models, and the lattice-based coarse-grained Side CHain Only (SICHO) model. It is found that the CHARMM19 and SICHO models lead to initially rapid refinement for a set consisting of 11 targets from past critical assessment of protein structure prediction (CASP) competitions, but they are eventually outperformed by the CHARMM22 and PRIMO models in consistently reaching near-native conformations over the course of 100 refinement cycles.