Estimation of coordinate uncertainties

Due to the large number of observables in an atomic resolution refinement, the inversion of the normal matrix of the refinement can be used to estimate standard uncertainties for the refined parameters (Cruickshank, 1970 Press et al., 2(X)2). The corresponding calculations are enormous and until recently could only performed on large mainframe-type computers. However, with the rapid progress in computing technology, the matrix inversion can now be done on standard crystallographic workstations in a couple of hours.  

For many stractures of macromolecules at atomic resolution presented in the literature, aU restraints were removed from the target function before the inversion of the normal matrix. However, recently it has been shown that in regions of the model where the restraints are important, for example when two sites belonging to two different conformations share the same electron density (Konig et al., 2003), the estimated standard uncertainties can be overestimated. Nevertheless, the estimated standard uncertainties for well-separated sites will be rehable. 

A structure at atomic resolution gives a much more detailed picture of a macromolecule than for example a structure at medium resolution. On one hand the higher resolution leads to smaller details becoming visible in electron density maps. On the other hand the much more elaborate parameterization of the model allows answering qualitatively new questions, like whether the anisotropic ellipsoids of two atoms in an active site are pointing towards each other. 

Before going into interpretation of the structural data, the overall quality of the model needs to be assessed. In addition to the standard statistics provided for models at lower resolution (Rwork. free. aii. agreement of the model with stereochemical restraints, mean B-values, etc.), quantities that characterize the atomic resolution model such as agreement of the anisotropic displacement parameters with the restraints imposed, must be quoted. Some interesting statistics on ADPs can be obtained with Ethan Merritt s PARVATI-server (Merritt, 1999). 

Due to the strength of the diffraction data, structures at atomic resolution may have more incidences of abnormal values as normally expected by the validation 

---