Case-controlled structure validation

Abstract

Although many factors influence the quality of a macromolecular crystal structure, validation criteria are usually only calibrated using one of these factors, the resolution. For many purposes this is sufficient, but there are times when one wishes to compare one set of structures with another and the comparison may be invalidated by systematic differences between the sets in factors other than resolution. This problem can be circumvented by borrowing from medicine the idea of the case-matched control: each structure of interest is matched with a control structure that has similar values for all relevant factors considered in this study. In addition to resolution, these include the size of the structure (as measured by the volume of the asymmetric unit) and the year of deposition. This approach has been applied to address two questions: whether structures from structural genomics efforts reach the same level of quality as structures from traditional sources and whether the impact factor of the journal in which a structure is published correlates with structure quality. In both cases, once factors influencing quality have been controlled in the comparison, there is little evidence for a systematic difference in quality.

Figure 1

Variation in mean validation criteria over time, comparing the structures deposited from 1972 to 1985 with matched structures deposited over subsequent periods. Error bars in this and subsequent figures indicate the estimated standard uncertainty of the mean, computed as the sample standard deviation divided by the square root of the number of observations. (a) Overall average G value computed by PROCHECK (Laskowski et al., 1993 ▶). (b) Percentage Ramachandran outliers determined using the criteria established by Kleywegt & Jones (1996 ▶).

Figure 2

Variation in validation criteria with resolution, comparing structures with a resolution limit from 2.05 to 2.06 Å with matched structures in other resolution ranges. (a) Histogram of reported resolution limits over the set of structures considered in this study. There are 33 structures with a resolution lower than 5 Å, but these would not appear on the scale of this histogram. (b) R _work (solid line) and R _free (dashed line). (c) Percentage of residues in favoured regions of the Ramachandran plot, as assessed by MolProbity (Davis et al., 2007 ▶).

Figure 3

Variation in validation criteria with asymmetric unit volume, comparing structures with asymmetric unit volumes between 100 000 and 102 000 ų with matched structures in other volume ranges. (a) Percentage of residues in favoured regions of the Ramachandran plot, as assessed by MolProbity (Davis et al., 2007 ▶). (b) R _free.

Figure 4

Scatter plot showing R _free as a function of resolution. Red points represent structures published in ‘high-impact’ journals, while black points represent structures published elsewhere.