The last step in the determination of a macromolecular structure is refinement. Refinement is the process whereby the model is altered to best match the facts known about the structure. Our knowledge about a structure will usually come from the diffraction of X-rays from a crystal and the facts of stereochemistry -- the bond lengths, bond angles, etc. -- expected for this type of molecule. Occasionally we will have other sources of information: experimental phases, similarity to a known structure, self similarity, or neutron diffraction. The final model should be consistent with everything known about the structure.
The work of refinement is divided between the computer and the crystallographer. The computer is very good at fine tuning a model to match the observations while the crystallographer is good at recognizing gross errors in the model and building entirely new solutions. The purpose of the TNT package of programs is to perform the automated optimization of the model and provide tools for use during the model rebuilding phase.
TNT does not attempt to correct gross errors in a model. It will not move atoms out of density to place them in ``better" density nor will it add or delete atoms. TNT will alter the parameters of the model to improve its fit to the observations and it will provide clues as to the location of gross errors.
In TNT a model is stored in a coordinate file (whose named ends in `.cor'). The structure of the model, the sequence and connectivity of all chains, is described in a sequence file (`.seq'). The details of the refinement -- the identity of the restraints, their weights, and which parameters to vary and how to vary them -- are stored in a control file (`.tnt'). Most of refinement is performed by editing the control file and executing the tnt command.
After a series of refinement cycles you should examine your model along with the current 2Fo-Fc and Fo-Fc maps (generated by make_maps). Each residue should be examined with particular attention paid to those residues which are most likely to be in error.
When the model is optimized to match the observations the observations are treated as restraints. While the overall agreement between the model and its restraints might be quite good, some individual atoms might agree badly. If the observations were measured without error and the type of model is sufficient to describe a molecule the model should be able to agree with all the observations at the same time. If some observations are not met by the model trouble is indicated. Portions of the model which do not fit the observations should be examined carefully on a graphics system.
TNT will provide you with lists of the atoms which are in greatest violation of the restraints. It will list the worst examples of bond lengths, bond angles, etc. with the geometry command. The ncs command will list the worst violations of the noncrystallographic symmetry, and the dscreen command will list the atoms which lie nearest to features in the difference map. These lists are indicators of the trouble spots in the model.
A separate type of error is the omission of atoms. Places where the model should contain atoms, but does not, can be found by looking for peaks in the Fo-Fc map. The pekpik command will generate marker atoms at the site of each of the highest 30 peaks in the difference map. These sites might be the location of alternative side chain conformations, the binding of solvent (either water or something else). While you should also examine each peak and try to interpret its cause there is no need to place atoms in every peak. Some peaks simply cannot be explained.