A statistical analysis is reported of experimental data and coordinate
s of a set of 97 NMR structures deposited in the PDB. The aim is to as
sess the quality of these structures in relation to the amount of expe
rimental information. Experimental restraints were analysed using the
program AQUA. Many nomenclature inconsistencies between deposited rest
raint and coordinate files were observed. The experimental restraint f
iles were found to contain a high proportion of redundant restraints.
Procedures for analysing and correcting the inconsistencies and restra
int counts are described. The analysis of NOE restraint violations (us
ing AQUA) and of a wide variety of geometrical quality indicators (usi
ng PROCHECK-NMR and WHAT IF) provides a reference for other NMR struct
ure determinations. The extent of NOE violations is anti-correlated wi
th the quality of the Ramachandran map. The precision as measured by t
he circular variance of backbone dihedral angles, does increase with t
he amount of experimental data, as expected, but is sometimes overesti
mated. Bond lengths, bond angles and planarity of groups can deviate c
onsiderably from ideal values. Outliers appear to cluster per laborato
ry, indicating that the results depend on particulars of refinement pr
otocols and/or software. We have identified a problem of atom overlap
in a number of refined structures. We recommend adhering to the standa
rd nomenclature as put forward by an IUPAC Task Group, to ensure consi
stency between restraints and coordinates, and to omit redundant restr
aints from the deposition. The results obtained from this analysis and
the AQUA program are available through the World Wide Web. (C) 1998 A
cademic Press