Q-score as a reliability measure for protein, nucleic acid and small-molecule atomic coordinate models derived from 3DEM maps

Abstract

Atomic coordinate models are important for the interpretation of 3D maps produced with cryoEM and cryoET (3D electron microscopy; 3DEM). In addition to visual inspection of such maps and models, quantitative metrics can inform about the reliability of the atomic coordinates, in particular how well the model is supported by the experimentally determined 3DEM map. A recently introduced metric, Q-score, was shown to correlate well with the reported resolution of the map for well fitted models. Here, we present new statistical analyses of Q-score based on its application to ∼10 000 maps and models archived in the EMDB (Electron Microscopy Data Bank) and PDB (Protein Data Bank). Further, we introduce two new metrics based on Q-score to represent each map and model relative to all entries in the EMDB and those with similar resolution. We explore through illustrative examples of proteins, nucleic acids and small molecules how Q-scores can indicate whether the atomic coordinates are well fitted to 3DEM maps and also whether some parts of a map may be poorly resolved due to factors such as molecular flexibility, radiation damage and/or conformational heterogeneity. These examples and statistical analyses provide a basis for how Q-scores can be interpreted effectively in order to evaluate 3DEM maps and atomic coordinate models prior to publication and archiving.

Keywords: B factors; Q-scores; cryoEM; structure; validation.

Figure 1

Relationship between Q-score and reported resolution, d, using EMDB maps and their associated atomic models in the PDB. (a) A plot showing each map and model pair as a filled circle, with a dotted line showing a regression using a third-degree polynomial. (b–f) Side chains at various resolutions, with corresponding decreasing Q-scores, averaged over the whole model (Q_model) or averaged over the residue shown (Q_residue). (g–i) α-Helices at three different resolutions between 5 and 10 Å.

Figure 2

(a) Plot of Q-scores versus reported resolution for ∼10 000 maps and models in the EMDB (the same data set as in Fig. 1 ▸). The dotted curves above and below the Q_peak curve enclose 95% of the data points (equations 2, 3 and 4). (b–e) Illustration of maps and models with Q-scores outside the 95% curves. Overall Q-scores for each model are indicated with Q_model and are colored red if outside the 95% curves. In (b) and (d) the Q-scores are inside the 95% curves after properly fitting the model to the map and re-calculating the Q-scores.

Figure 3

Examples of Q-score application in proteins and in nucleic acids. (a) β-Galactosidase protein complex with (b) per-residue backbone and side-chain Q-scores; example residues with Q-scores marked on the plot are marked (i) and (ii). (c) Ion-channel protein complex; one of the two proteins in the complex is shown in (d), with a ribbon display color-coded by residue Q-score. (e) Per-residue backbone and side-chain Q-scores for one ion-channel complex protein; an area with low Q-scores is marked (iii). (f) RNA-only Tetrahymena ribozyme; the ribbon model is color-coded by nucleotide Q-score. (g) Q-scores of phosphate, sugar and base atoms in each nucleotide; Q-scores for three residues which are well resolved are shown in (v) and an area with low nucleotide Q-scores is marked (iv).

Figure 4

Application of Q-scores to small molecules. (a) Segmented 3DEM map of coronavirus NL63 spike proteins (blue, orange, green) with Asn-associated glycans (yellow). (b, c) Two example glycans, with Q-scores for each component saccharide. (d) The Q-scores of each saccharide are plotted. (e, h) Two 3DEM maps of β-galactosidase with the same reported resolution of 1.9 Å. Two models of the ligand PTQ and three interacting protein residues, along with Q-scores, are shown in (f) and (g) for the map in (e) and in (i) and (j) for the map in (h).

Figure 5

Atomic B factors from Q-scores. Top row: two residues in four different 3DEM maps and models with resolutions of ∼1 to ∼4 Å. Second row: model maps generated with atomic B factors calculated by scaling Q-scores. Third row: model maps generated with atomic B factors set to 0. Fourth row: bar plots of CC-mean (cross-correlation about the mean) between the 3DEM map and model maps generated with atomic B factors calculated using a range of scaling factors (0–300); the bar with the highest CC-mean value is colored orange.