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. As a dataset of decoys to compare with real protein structures, we studied submissions to the bi-annual CASP competition (specifically CASP11, 12, and 13), in which researchers attempt to predict the structure of a protein only knowing its amino acid sequence. Our analysis reveals that many of the submissions possess cores that do not recapitulate the features that define real proteins. 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 deep learning method, 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 decoys from the CASP competitions equally well, if not better than, state-of-the-art methods that incorporate many additional features.
Download