The course of refinement will depend upon the errors in the model. When a structure is solved by Molecular Replacement the relative position of the atoms are usually quite good but their absolute position could be in error by a great deal. This is because the rotation function and translation function solutions could be in error by several degrees and maybe an Ångstrom. These are very large errors and normally could not be corrected except for the strong phasing power of the essentially correct core of the molecule. This situation calls for rigid-body refinement, the motion of the core of the molecule as a single unit. In addition, the domains of the molecule might have shifted in the current molecule relative to those of the prototype. Just to be sure, after the overall rigid-body refinement the molecule should be broken down into rigid domains and refined again.
Usually rigid-body refinement is performed using only the data to 4.0 Ångstrom resolution.
When a structure is solved by Molecular Modification, the structure of a molecule which is a modified form of a known structure in an isomorphous crystal, often the cell constants will have changed. This change will require that refinement begin with rigid-body refinement as well. In this case the goal is to principally translate the molecule not to rotate it but the method is the same. If the modification causes a change in the orientation of the domains, rigid-body domain refinement will be required as well.
Whenever performing rigid-body refinement you should consider setting the weight for nonbonded contacts to zero. In a Molecular Replacement model the loops on the surface are quite likely misplaced. Forcing the center of the molecule to be displaced because of a surface bad contact might not be a good idea.
Models derived from these methods will be quite good once the overall corrections to orientation have been made. One may start immediately using the higher resolution data because the expected size of the shifts required to correct the model is small. With a Molecular Replacement model you may want to start at 2.5Å\ resolution while a Molecular Modification model probably can start at the limit of the data collected.
When the initial model has been constructed by building into a map, such as in MIR and MAD structure determination, overall adjustments in orientation will not be required. However such models will have many misplaced atoms and probably many places where the agreement with the geometric restraints is poor. Because large shifts will be required to correct the model only the lower resolution data should be used at first. The best strategy appears to be to start with the resolution limit of your map. A round of loosening up the geometry and tightening down should be performed. (This step will be discussed in more detail later.) It might pay to loosen and tighten a second time before extending resolution.
If the model improved a great deal during refinement you should enter model building without adding more data. If the refinement did not change the model much, and you think it fits the data you are using well enough, you should extend the resolution and run more refinement. If the agreement between your model and your data cannot be made satisfactory it is likely that something is seriously wrong. Either the model or the data must be corrected before preceding.
Rigid-body refinement is performed by placing one or more COMBINE statements in your control file, fixing the B factors and occupancies with two CONSTANT statements, and running the refinement by typing ``tnt 1 5 sd''.
To run free atom refinement simply remove the COMBINE statement(s), remove the ``CONSTANT B'' card and type ``tnt 1 30''. Of course you may use whatever cycle numbers you want.
There are two questions left to be answered. How many cycles should I run and what weights should I use? The first question is easy. About 30 cycles seems to get the atoms with low temperature factors to where they are going. Additional cycles will continue to move atoms, in some cases quite a bit, but those atoms will be the ones with very high B values and their position is very imprecisely defined by the diffraction data anyway.