RosettaHoles2: a volumetric packing measure for protein structure refinement and validation

Abstract

We present an improved version of RosettaHoles, a methodology for quantitative and visual characterization of protein core packing. RosettaHoles2 features a packing measure more rapidly computable, accurate and physically transparent, as well as a new validation score intended for structures submitted to the Protein Data Bank. The differential packing measure is parameterized to maximize the gap between computationally generated and experimentally determined X-ray structures, and can be used in refinement of protein structure models. The parameters of the model provide insight into components missing in current force fields, and the validation score gives an upper bound on the X-ray resolution of Protein Data Bank structures; a crystal structure should have a validation score as good as or better than its resolution.

Figure 1

The complete set of parameters of the RosettaHoles2 score. The x axis is shown on a logarithmic scale for clarity. The first category (top panel) includes hydrophobic side chain atoms CH₁, CH₂, CH₃ and hydrophobic backbone atoms in loops. The second category (second panel) consists of polar side chain atoms and polar backbone atoms in loop configurations. The third group of atoms comprises the four backbone atom types in helix or sheet configurations (third panel). The fourth group of atoms, proline backbone nitrogen, sulfur, aromatic carbons, and the outermost sidechain carbon on glutamine and asparagine, do not cluster with the others. The contribution to the RosettaHoles2 score for an atom is simply the dot product of the binned atomic volumes with the relevant parameters above. The parameter values shown are for the S_DECOY score; the parameters of the S_Val and S_RESL models follow similar trends.

Figure 2

Dependence of S_VAL, S_RESL, and S_DECOY on the extent of structural flaws. Starting from two structures of T4 lysozyme (177L and 179L), which are nearly identical except for the packing flaws in 179L, a spectrum of intermediate structures was generated ranging from good packing (left side, 177L) to very bad packing (right side, 179L). The sensitive S_RESL score increases linearly with small flaws but plateaus as flaws become more severe, whereas the complementary S_DECOY score increases slowly with small flaws but quite rapidly as flaws become more severe. The combined S_VAL score increases roughly linearly with the magnitude of the packing flaws, from minor to severe.

Figure 3

S_VAL for crystal structures in the Protein Data Bank as of Summer 2009. Only structures that contain primarily protein, are greater than 10 kDa, and have little missing density are shown. The packing score correlates with resolution and is calibrated so that the S_RESL score should be lower than the resolution in most cases. Points above the dashed line in the figure deviate by more than 5 standard deviations from the population of structures of similar resolution. The most extreme of these outlying structures are labeled on the plot. On the right are histograms showing the distribution of RosettaHoles2 scores for structures at 1.5, 2.5, and 3.5 Å resolution.