Can Predicted Protein 3D Structures Provide Reliable Insights into whether Missense Variants Are Disease Associated?

Abstract

Knowledge of protein structure can be used to predict the phenotypic consequence of a missense variant. Since structural coverage of the human proteome can be roughly tripled to over 50% of the residues if homology-predicted structures are included in addition to experimentally determined coordinates, it is important to assess the reliability of using predicted models when analyzing missense variants. Accordingly, we assess whether a missense variant is structurally damaging by using experimental and predicted structures. We considered 606 experimental structures and show that 40% of the 1965 disease-associated missense variants analyzed have a structurally damaging change in the mutant structure. Only 11% of the 2134 neutral variants are structurally damaging. Importantly, similar results are obtained when 1052 structures predicted using Phyre2 algorithm were used, even when the model shares low (<40%) sequence identity to the template. Thus, structure-based analysis of the effects of missense variants can be effectively applied to homology models. Our in-house pipeline, Missense3D, for structurally assessing missense variants was made available at http://www.sbg.bio.ic.ac.uk/~missense3d.

Keywords: Phyre2 protein structure prediction; missense variants; protein structure prediction; structure-based prediction; variant effect prediction.

Graphical abstract

Fig. 1

Pipeline to analyze the structural impact of missense variants in experimental and predicted structures.

Fig. 2

Performance of structural analysis on experimental structures. For each feature, the TPR on disease-associated (Disease) and the FPR on neutral (Neutral) variants are plotted as bars. The ratios of the true-positive to false-positive rates (TPR/FPR) are given, and for ease of viewing, these are connected by a line. The overall TPR and FPR on the entire data set are also reported. Significance at P < 0.01 (denoted by **) is evaluated in a one-tailed test of the difference of two proportions. Panel a are the results from the set of high-quality X-ray structures (resolution < 2.0 Å) from MolProbity's Top8000 database . Panel b are the results on a second independent data set of 855 structures with lower resolution (< 2.5 Å) from the PDB.

Fig. 3

The distribution of the RMSD (Å) between the Phyre2-predicted models and the true experimental coordinates binned according to the % sequence identity between query and template. The central line in a box is the median RMSD with its value reported. The upper and lower box boundaries are the upper and lower quartiles. The whiskers extend up an additional 1.5 × the difference between the median and the upper quartile and down an additional 1.5 × the difference between the median and the lower quartile.

Fig. 4

Performance of structural analysis on predicted models at different sequence identities. For each sequence identity bin, the fractions of positive predictions for the disease-associated and neutral variants are shown for both the predicted and the corresponding experimental structures. 95% Confidence intervals on the positive rates are shown as lines.

Fig. 5

Histogram of the relative frequencies of RMSD (Å) of predicted models grouped according to whether the variant was a TP and FP. The bin labeling shows the upper bound; for example, 0.5 denotes the range 0.0 Å ≤ RMSD < 0.5 Å. The last bin is RMSD ≥ 6 Å. The relative percentage displayed on the Y axis is the fraction of true positives and similarly the fraction of false positives in each bin range.

Fig. 6

Structural analysis of p.His107Tyr in the experimental and predicted structures of carbonic anhydrase (CA2). (a) His 107 in the wild-type (WT) (PDB ID: 2FOS), (b) Tyr 107 in the mutant (MUTANT) modeled from the WT, (c) Phyre2-predicted structure of the wild type with His 107 (PREDICTED WILD TYPE) and (d) predicted structure of the Tyr 107 mutant based on the Phyre2-predicted wild-type structure (PREDICTED MUTANT). In all four panels, the Cα traces of the structures analyzed by Missense3D are shown in gray. In the predicted structures (c and d), the Cα trace and side-chain positions shown in panels a and b are shown in pink. The side chains of the His 107, Tyr 107 and Glu 117 are show by chemical type (green for non-polar, blue for positive and red for negative). Figures were generated using PyMOL .

Fig. 7

Structural analysis of variant p.Cys52Arg in the crystal and predicted structures of alpha-galactosidase. (a) Cys 52 in the wild type, (b) Arg 52 in the mutant modeled from the wild type, (c) Phyre2 wild-type predicted structure of the Cys 52 using a model with RMSD to the crystal structure of 1.4 Å (PREDICTED WILD TYPE) and (d) Phyre2-predicted wild-type structure of the Cys 52 using a model with RMSD to the crystal structure of 2.8 Å (PREDICTED WILD TYPE (2)). Color scheme for the four panels as in Fig. 7. In addition, the SG atoms are shown in yellow.