Tips and Tricks in Crystallography - Structure Refinement

Mariusz Jaskolski: Sterechemical Restraint Libraries for Nucleic Acids

As part of a project aimed at revising the stereochemical restraint libraries for nucleic acids, a server, RestraintLib, has been created. It prepares external (custom) restraint files for the phosphate groups in DNA and RNA structures (submitted as PDB files). At present, the output is prepared for Refmac5, but other formats will be added as well.

The discussion of the new restraints and the method of their generation are presented in Nucleic Acids Research Advanced Access published August 12, 2016.

Mariusz Jaskolski: Geometrical Restraints of His Residues
A recent paper, already listed on this bulletin board, will give you a hint about improving the geometrical restraints of His residues, depending on their protonation/tautomeric form, especially at high resolution:

M.Malinska, M.Dauter, M.Kowiel, M.Jaskolski, Z.Dauter (2015) Protonation and geometry of histidine rings. Acta Cryst. D71, 1444-1454.

Xinhua Ji (NCI): Phenix Tool For Omit Map Calculation
Phenix.polder calculates omit maps for selected atoms while preventing the bulk solvent mask to 
flood into the selected area and its vicinity. The tool can be useful in cases where the density of selected
atoms is weak and possibly obscured by bulk solvent.

Usage
phenix.polder  model.pdb  data.mtz  selection='chain A and resseq 123'

Examples
https://www.phenix-online.org/presentations/phenix_polder.pdf

Documentation
http://www.phenix-online.org/version_docs/dev-2356/reference/polder.html

Mariusz Jaskolski: Special Positions. A recent PDB deposit from the CBB (returned 
with comments about errors) has prompted me to sensitize those of us who are less 
versed in crystallography about the significance of special positions and their 
consequence on refined parameters. (to read on, click here)

Pavel V. Afonine, Ralf W. Grosse-Kunstleve, Peter H. zwart, Thomas C. Terwilliger, Nigel W. Moriarty & Paul D. Adams: Phenix.refine: Crystallographic Structure Refinement in PHENIX. A combination of highly efficient programming tools and new or improved crystallographic algorithms provides a very high level of automation and robustness in structure refinement. Their implementatiion in phenix.refine -- state-of-the-art refinement module of PHENIX pachage -- resulted in a complete set of tolls that cover most of refinement needs and scenarios, such as: ... (full article).

Pavel V. Aftonine's recent lecture on phenix.refine (click here).

Dr. Mohana Rao (NCI): "Headache" Medicines. During refinement, one residue or another has some problems in its placement. To some extent, this becomes evident in the R-list that is computed when the maps are made using the CNS program. Once I wrote a program that checks each atom in the PDB file against the density and flags off atoms that are dubious in their placement. After this is done this program tries to account for residual densities [either in (2Fo-Fc) or (Fo-Fc) maps] and to locate water molecules by distance criteria. It writes a PDB file for the water molecules including estimated temperature factors. This program, in fact, was developed long before the present programs came into vogue.

Dr. Alexander Zdanov (NCI): R and Rfree. A couple of our group meetings ago, we had a short discussion on what could be an acceptable difference or ratio between Rfree and R during crystallographic refinement; however, it was left somewhat unfinished since the time ran out. I think that this is an important issue and we could share our views and tips so everyone will gain something. The concept of Rfree was introduced in 1992 by Axel Brünger and was almost immediately accepted by protein crystallographers. Rfree is the same as R but is calculated only for a test set of reflections (5-10% of the total number of reflections) excluded from the refinement. The assumption is that test reflections are randomly selected and have only errors, which do not correlate with those of the rest of the data included in the refinement. When Rfree was introduced, it was only calculated for the pleasure of the individual carrying out the refinement, and only later it was directly incorporated into the refinement for cross-validations. Any maximum likelihood target requires that you have selected a test set of reflections. It is important that the reflections should be good enough, that is they are random and so on, and also the number of the reflections is appropriate. Therefore, it is not wise to reduce the number of reflections in the test set just to try to decrees Rfree value. I usually choose (by default) 10% or so, but at least about 1000 reflections. The Rfree/R ratio has been analyzed in “Rfree and the Rfree ratio. I. Derivation of …”, by I.J. Tickle, R.A. Laskowski, and D.S. Moss, Acta Crystallogr. (1998), D54, 547-557. I like this paper, it has good theoretical background and very clear statistical diagrams on the dependence of Rfree/R ratio from the Na/f ratio (Na - is number of atoms, f - is number of reflections) at different resolutions and different number of refined parameters per atom, based on the 725 macromolecular structures deposited in the PDB. Most of the structures refined with four parameters per atom (x,y,z and individual isotropic B) at the resolutions 2.5-1.5A fall in the area 1.1-1.5, that is for the R=18%, the corresponding Rfree would be 19.8-27%. It is also important to remember that Rfree is a measure of how well your current model agrees with diffraction data in the test set, and that the model does not know the existence of the test set of reflections. In other words, if the model for whatever reasons lacks some parts of the structure, which contributed to the diffraction, Rfree/R inevitably must be higher.