Using physical features of protein core packing to distinguish real proteins from decoys

Abstract

The ability to consistently distinguish real protein structures from computationally generated model decoys is not yet a solved problem. One route to distinguish real protein structures from decoys is to delineate the important physical features that specify a real protein. For example, it has long been appreciated that the hydrophobic cores of proteins contribute significantly to their stability. We used two sources to obtain datasets of decoys to compare with real protein structures: submissions to the biennial Critical Assessment of protein Structure Prediction competition, in which researchers attempt to predict the structure of a protein only knowing its amino acid sequence, and also decoys generated by 3DRobot, which have user‐specified global root‐mean‐squared deviations from experimentally determined structures. Our analysis revealed that both sets of decoys possess cores that do not recapitulate the key features that define real protein cores. In particular, the model structures appear more densely packed (because of energetically unfavorable atomic overlaps), contain too few residues in the core, and have improper distributions of hydrophobic residues throughout the structure. Based on these observations, we developed a feed‐forward neural network, which incorporates key physical features of protein cores, to predict how well a computational model recapitulates the real protein structure without knowledge of the structure of the target sequence. By identifying the important features of protein structure, our method is able to rank decoy structures with similar accuracy to that obtained by state‐of‐the‐art methods that incorporate many additional features. The small number of physical features makes our model interpretable, emphasizing the importance of protein packing and hydrophobicity in protein structure prediction.

Core packing of well-defined x-ray and NMR structures is the same

“…In summary, when we analyze all structures in the PDB determined by NMR, for which restraint data is available, they on average possess smaller, more densely packed cores compared to high-resolution x-ray crystal structures. Interestingly, smaller, but more densely packed cores have also been found in many computationally generated low-scoring structures with incorrect backbone placement submitted to the Critical Assessment of protein Structure Prediction (CASP) competitions 14 .…”

“…In previous work, we analyzed a non-redundant set of high-resolution x-ray crystal structures (with resolution < 1.8 Å) and found that the cores (residues with zero solvent accessible surface area) of these proteins represent about 8% of the total number of residues in the protein. In addition, the packing fraction, φ, in protein cores is ∼ 0.55 ± 0.01 13,14 . In earlier work, using a subset of NMR structures with a large number of distance restraints per residue, we concluded that the cores of NMR-determined protein structures had a higher packing fraction than those of high-resolution protein structures determined by x-ray crystallography 12 .…”

Core packing of well‐defined X‐ray and NMR structures is the same

“…Interestingly, smaller, but more densely packed cores have also been found in many computationally generated lowscoring structures with incorrect backbone placement submitted to the Critical Assessment of protein Structure Prediction (CASP) competitions. 26 To investigate the discrepancies in the core packing properties in greater detail, we compared protein structures that had been determined by both X-ray crystallography and NMR. We assembled a data set of 702 pairs with both a high-resolution X-ray crystal structure (<2.0 Å) and an NMR structure with available restraints, where the pairs have >90% sequence similarity.…”

Current MD forcefields fail to capture key features of protein structure and fluctuations: A case study of cyclophilin A and T4 lysozyme

“…However, one must be able to identify when the REMD simulation is close to the experimental structure, without knowing the experimental structure beforehand. The distance to the x-ray crystal structure can be estimated using protein decoy detection methods [14,[21][22][23], which have their own limitations, and would limit the ability of REMD to identify experimental conformations. It is likely that the CHARMM36m forcefield possesses low-lying energy minima that are distinct from those sampled by x-ray crystal structures.…”

“…The algorithm takes z-slices of the protein, determines the solvent accessibility of the sets of contours using a probe molecule of a given radius, and integrates the SASA over the slices. We use a water-molecule-sized probe with radius 1.4 Å and z-slices with thickness ∆z = 10 −3 Å, which were used in previous work [11,14,34,48,49]. To normalize the SASA, we take the ratio of the SASA within the context of the protein (SASA context ) and the SASA of the same residue X extracted from the protein structure as a dipeptide (Gly-X-Gly) with the same backbone and side-chain dihedral angles:…”