The structures obtained from homology modeling methods are of intermediate resolution 1-3Ã… from true structure. Energy minimization methods allow us to refine the proteins and obtain native like structures. Previous work shows that some of these methods performed well on soluble proteins. So we extended this work on membrane proteins. Prediction of membrane protein structures is a particularly important, since they are important biological drug targets, and since their number is vanishingly small, as a result of the inherent difficulties in working with these molecules experimentally. Hence there is a pressing need for alternative computational protein structure prediction methods. This work tests the ability of common molecular mechanics potential functions (AMBER99/03) and a hybrid knowledge-based potential function (KB_0.1) to refine near-native structures of membrane proteins in vacuo. A web based utility for protein refinement has been developed and deployed based on the KB_0.1 potential to refine proteins.

<div>X-ray crystallography is a foundation of the modern structural biology. Thus, refinement of crystallographic structures remains an important and actively pursued area of research. We have built a software solution for refinement of crystallographic protein structures using X-ray diffraction data in conjunction with state-of-the-art MD modeling setup. This solution was implemented on the platform of Amber 16 biomolecular simulation package, making use of graphical processing unit (GPU) computing. The proposed refinement protocol consists of a short MD simulation, which represents an entire crystal unit cell containing multiple protein molecules and interstitial solvent. The simulation is guided by crystallographic restraints based on experimental structure factors, as well as conventional force-field terms. We assessed the performance of this new protocol against various refinement procedures based on the Phenix engine, which represents the current industry standard. The evaluation was conducted on a set of 84 protein structures with different realizations of initial models; the main criterion of success was free R-factor, R_free. Initially, we performed the re-refinement of the models deposited in the PDB bank. We found that in 58% of all cases our protocol achieved better R_free than Phenix. As a next step, we conducted the refinement on three different sets of lower-quality models that were manufactured specifically to test the competing algorithms (average C^α RMSD from the target structures 0.75, 0.89, and 1.02 Å). In these tests, our protocol outperformed the refinement procedures available in Phenix in up to 89% of all cases. Aside from R-factors, we also compared geometric qualities of the models as measured by MolProbity scores. It was found that our protocol led to consistently better geometries in all of the refinement comparisons.</div><div>Recently, a number of attempts have been made to fully utilize the information encoded in protein diffraction data, including diffuse scattering, which is dependent on molecular dynamics in the crystal. To understand the nature of this dependence, we have chosen three different crystalline forms of ubiquitin. By post-processing the MD data, we separated the effects from different types of motion on the diffuse scattering profiles. This analysis failed to identify any features of the diffuse scattering profiles that could be uniquely linked to certain specific motional modes (e.g. small-amplitude rocking motion of protein molecules in the crystal lattice). However, we were able to confirm the previous experimental observations, made in the combined X-ray diffraction and NMR study, suggesting that the amount of motion in the specific crystal is reflected in the amplitude of diffuse scattering.</div>