The accuracy of NMR protein structures in the Protein Data Bank

We recently described a method, ANSURR, for measuring the accuracy of NMR protein structures. It is based on comparing residue-specific measures of rigidity from backbone chemical shifts via the random coil index, and from structures. Here, we report the use of ANSURR to analyse NMR ensembles within the Protein Data Bank (PDB). NMR structures cover a wide range of accuracy, which improved over time until about 2005, since when accuracy has not improved. Most structures have accurate secondary structure, but are too floppy, particularly in loops. There is a need for more experimental restraints in loops. The best current accuracy measures are Ramachandran distribution and number of NOE restraints per residue. The precision of structure ensembles correlates with accuracy, as does the number of hydrogen bond restraints per residue. If a structure contains additional components (such as additional polypeptide chains or ligands), then their inclusion improves accuracy. Analysis of over 7000 PDB NMR ensembles is available via our website ansurr.com.