Likewise, the argument that disordered atoms will inflate the B factors of the ordered atoms (like James Irving mentioned) is valid the other way around too: an ordered C-beta will keep the B factor for C-gamma at bay, and so on. If the B factor is meant to reflect degree of disorder, then a mutual effect is completely desirable. However, we all know that B factors are not perfect, either.
Wouldn't this mutual compensation effect flatten out the side-chain B's of the offending residue, meaning that subsequent inspection by the end-user would be less likely to see a discrepancy in B factors? In addition, when building a structure we all have subjective idea based on the density and our chosen contour levels as to whether an atom is 'there' or not, but B factors aren't so clear cut.
If C-beta is well defined, we have a pretty good idea about where C-gamma is and so on. Contour the 2Fo-Fc map at 0.3 sigma and you will likely see some density.
Surely there will be a point where geometry terms dominate the positioning of the 'invisible' side-chain atoms, and they cease to be representative of the X-ray data...?
I think we should not dumb down our models because we fear someone else might not fully understand them. With 'dumbing down', I mean omitting useful information.
Absolutely - it is paramount that the accuracy and completeness of information not be compromised for the sake of contemporary non-crystallographers using contemporary visualisation software. There can always be improvements made to the software and the way users of a structural database are educated, but the deposited data will be, for the most part, frozen in time. It seems that a factor behind some of this controversy is the conventions imposed by the present PDF and mmCIF file formats. If there was a field in which atoms could be labelled as (in the opinion of the structure's author) consistent with electron density (true) or only inferred (false), then the argument over whether disordered atoms should be deleted or B-factored-out would be uncoupled from considerations about how best to accurately represent the data to end-users. Importantly, cases where side chains are modelled according to considerations other than the x-ray data would be readily identifiable. James
