 |
 |
I would echo Ethan on this metric being something of a relic and add a bit more data. Several years ago I tried to get a practical solution to the questions: - when is a refinement finished? - how to detect the correctable abnormalities (errors) in a structure so they can all be corrected during refinement and before deposition ? - what is the 'quality' of a structure? The idea was that our in-company structure refinements would give high quality structures (at least no gross errors) that could be used by chemists and the refinement work would not just go on forever. Anyway, if you consider a simple count of the number of residues with an identifiable problem (say, a grossly distorted geometry, a disallowed phi-psi, a very unlikely chi-1 angle, very short contacts, a really poor correlation with electron density, all defined is rather generous ways) as sensible proxy for a useful structure 'quality' then coordinate accuracy quantities like DPI (or its older relation, the Luzzati plot) turned out to be pretty much useless. For that matter, like the current PDB we initially used SFCHECK and PROCHECK in our standard validation system but ultimately found them not really robust enough for all cases and switched to direct use of REFMAC with 0 refinement cycles as a structure factor/map calculator and applied the Richardson's phi- psi data to create more accurate Ramachandran metrics.