Checking your imagination: applications of the free R value

Abstract

In 1990, Brändén and Jones [1] suggested that the process of electron-density map interpretation involves a degree of subjectivity on the part of the crystallographer. In the worst case, serious errors may be introduced into a model. In particular, their commentary stimulated the development of many different approaches for assessing the quality of protein crystal structures. Many of these methods validate models using statistical surveys of high-resolution X-ray structures (either of small molecules or of macromolecules) from which expected values or ranges of values are calculated for structural features. However, without the use of experimental diffraction data, one can never be sure whether a statistical ‘outlier’ (for example, in a Ramachandran plot) represents an error in the model or a genuine feature of the crystal structure [2], and the distinction between the two is obviously important. The ambiguity can only be resolved if the diffraction data is used in the quality assessment. Therefore, methods to assess if the model is an accurate representation of the experimental data must rely on statistics that involve both the diffraction data and the atomic model. Furthermore, few statistical surveys have been performed for macromolecular structures other than proteins, such as RNA, which means that knowledge-based rules are not available for the latter.