Why is the anisotropic B factor refined at low resolution? For example, PDBid:6Z32 (resolution is 3.4A) contains anisotropic B-factors. The counterpart in the database PDB-REDO has only isotropic B-factors and a lower R-factor.
Dear Jure,
Most likely you are mixing up different meanings of numbers in the PDB representations. The anisotropy in 6Z32 does not represent U tensor of individual ellipsoid displacement but rather and integrated TLS representation. This structure is obviously refined using TLS rigid body tensor approximations as is apparent from the coordinates file. If these are integrated into individual Bs then the result is expressed as anisotropic temperature factors. This is not "real anisotropic" refinement as the number of data points (reflections) does not allow for a meaningful anisotropic representation. Some old pieces of methodological work showed that stability of crystallographic refinement os conditioned on having approximately 4 data points (reflections) per every degree of freedom in the refinement (which translates into x,y,z,q,b11,b12,....b33). So as Murshudov showed when he presented his improved Refmac that one can refine anisotropy at almost any resolution but needs stronger and stronger restraints. All above mentioned conditions lead to at least two recommendations. (1) Depending on many conditions true real anisotropic refinement can be done at higher resolution than ~1.6Å. (2) Even if we have enough data points (reflections) we always start with isotropic refinement then try to model everything we can and only then introduce anisotropy as the final resort. This process is always is an experimental computational process as the space group, completeness, quality as measured by Rmerge or CC1/2, anisotropy of the data and many other factors influence how successful the anisotropic B refinement would be. Hence the appropriate resolution has to be established by success of the anisotropic refinement.
In presence of these difficulties it became a common practice to introduce TLS refinement as a useful tool to lower R factor and present fragmented mobility in low resolution structures. In my modest opinion this is widely abused and sometimes obscuring obvious capabilities to improve the structures by sucking the errors into the TLS factors. TLS introduced many years ago in classic works of Trueblood stand for Translation, Libration, and Screw tensors. As can be invoked from the names T and L tensors are symmetric as they represent three-dimensional decomposition of translations and rotational degrees of freedom. The S tensor is usually not symmetric as it correlates both T and L to obtain a physical displacement represented as U tensor of anisotropy of individual atoms. For more info I refer you to works of Merritt and Painter, and CCP4 programs.
Good luck.
Bog
Dear Boguslaw,
first of all thank you for your answer and explanation. Yes, I confused the values of TLS and B-anisotrpic (the entry in the PDB file is the same ANISOU). The data from the REDO database for this particular model shows no use of TLS, while Rwork is lower compared to the PDB entry.
Jure,
One additional remark. Do not trust PDBRedo. In my hands it never corrected more severe errors. I tried to communicate these findings to the principals like Berman or Terwilliger but never got anywhere. All structures in the PDB contain some kind of errors. They are larger or smaller but every structure (including mine) can be improved upon more detailed refinement. Particularly many errors are made in small conformational errors of side chains, incomplete water/solvent structure and in identity of many ligands. Many papers were written on these individual topics.
Good Luck.
Bog
- Badges
- Science topic
- Similar topics
- Engineering
- Materials Engineering
- Powders