Crystallographic refinement

Refinement is defined as optimization of all atomic model coordinates and B-factors against a kind of a map derived from experimental data. Besides the experimental target function also empirical force field parameters keep the model chemically reasonable.

After your refinement ".cmds" macros have been configured refine your model by entering the "REFINE" page and defining the "REF_XYZ", "REF_B", and "REF_OCCU" macros. (see MAIN_MENU:refine.html).

Pressing "e" in the image window EXITS minimizations after the completion of the current step.

Target functions

MAIN calculates positional and B-value from a series of target functions.

Six target functions are used:

The maximum likelihood refinement targets require use of TEST reflection set and R-free is mandatory.

All target function are scaled against the chemistry terms so that the ratio between target functions and chemistry terms is defined during the initail step and updated every 100 cycles. The scale is derived from the bond RMSD and the specified target. Default target is 0.02A.

> min adjust bond 0.02

Density target

This would be called refinement toward a map or real space refinement. In MAIN terminology "REFINE" differs from the "MINIMIZE" item, "REFINE" affects all atoms, whereas MINIMIZE optimizes positions of a selected, local group defined by the key "active" (MAIN_MENU:nice_sel.html).

When your model crystallographic R-value is still high in high 20-ties or higher, a refinement of your complete model toward a 2Fobs-Fcalc, 3Fobs-2Fcalc map or a similar (MIR, phased combined or density modified) can significantly improve your R-value.

Optimization of NCS operators for density averaging before applying cyclic averaging (which uses Fourier transformed maps) is a mandatory step.

Refinement against a difference (Fo-Fc)

It is one of the oldest forms of refinement. Nowadays it makes sense to apply it for multi crystal averaging and for high resolution structures as a short cut, since it is derivatives are based on a few points of electron density only.

Anisotropic overall correction

The matrix for anisotropic scaling of FOBS against FCALC is optimized during the "RE_PHASE" step assuming that the "ANISO" button in the "MAPS" calculation page is on.

Anisotropic correction term is optimized similarly as the bulk solvent correction by minimizing the Least Square or Maximum-likelyhood targets (depending if "ML_MAP" is on or off.)

Bulk solvent correction

The bulk solvent correction procedure is based on the flat model of bulk solvent method (Fokine and Urzhumtsev (2002) Acta Cryst. D58, 1387-1392) using optimized parameters. The parameters and envelope are derived during "RE_PHASE".

Turning it on by clicking the "BULK_SOL" in the "MAPS" caluclation page and then "RE_PHASE".

The bulk solvent correction term is optimized similarly as anisotropic correction by minimizing the LeastSquare or Maximum-likelyhood targets (depending if "ML_MAP" is on or off.)

Fast cooling versus slow cooling refinement

Random displacement of atomic positions and B-values from their current value is called "kicking". So instead of heating the system to 10 000K or more doing some simulation at these temperatures and then gradually lowering the temperature bath, the system is distorted in one step using a random number generator and then optimized. As refinement approaches its end, kick amplitudes are gradually decreased. Resulting initial geometry distortions can be so bigger than usually reached by high temperature dynamics. The distortions are removed in approximately 10 steps of minimization.

Kicking protocols are available through

Kicking allows allows B-value manipulation, whereas molecular dynamics allows only atomic coordinate treatment. (Clicking "KICK_B_5" kicks the B-values of atoms in the interval +/- 5.

 set coor kick 0.2 sele ... end
 set temp kick 5. sele ... end

Partial model refinement

A refinement procedure works well when (positional) derivatives are as correct as possible. They depend upon correctness of structure factor amplitudes and phases. Wrong phases result in meaningless positional gradients, which are in MAIN calculated from difference (Fobs-Fcalc) density maps. When model is incomplete the model structure factors are quite far away from their final form regarding both, amplitudes and phases.

The idea here is that we try to fulfill the partial model density with density information obtained from other sources. These can be an MIR map, a solvent flattened map, an averaged map, a 2Fobs-Fcalc map,

... actually any kind of map which contains additional density
information about the structure and is not included in the current partial structure model available. From this map the structure factors are calculated that guide the current atom models with help of a minimizer to their final place. This procedure is essentially a density modification method, but since it happens through MINIMIZE command it was placed under refinement. After refinement is completed you should look into your map MAP_FO, since it may contain substantial differences (improvements) when compared with other kind of maps (obtained by phase combination or density modification procedures).

The following procedure tries to reduce structure factor errors by combining partial model density map with MIR density map into a map used for structure factor calculation. From these Fcalc-s a difference (Fobs-Fcalc) map is calculated and later on used for model positional derivatives calculation within MINIMIZATION procedure.

The following procedure (dens_comb.cmds) requires 4 unit cell maps to be held in the memory (MAP_ORIG -the MIR map), a work space map (MAP_FOFC) and the (MAP_FO). It sets the variables for the whole run (

First define the energy term lists and set the crystallographic weights and some (maybe) undefined b-values.

> define impr by topo init
> set charge sele resi name ARG LYS ASP GLU .a .not -
> atom name C N CA O end   0.0
> set weigh sele atom name H* end 0.0
> set temp sele all end = 28.0

Copy the density of the MAP_ORIG map (MIR) into the MAP_FO and specify the number of shells used for the r-value calculation (scaling of Fcalc to Fobs).

> make map MAP_FO zero
> make map MAP_FO set -1 1 9999.
> make map MAP_FO from MAP_ORIG copy
> refl shel 15

Set the variables to control the density combination run: density energy scale factor (DENS_SCAL), number of minimization steps (MIN_STEP), number of minimization cycles after which a maps will be calculated, radius of the molecular model envelope (RAD_MASK), radius of the model map which will be considered to replace the current density, additional weight for the Fcalc map (WEIGHT_FC). The WEIGT_FC is an empirical factor. It should scale the model Fcalc map so that it does not disappear in the MIR map noise as well that this map would dominate over the MIR map. My criterion is to calculate the combined map so that the contours of the model fcalc map nicely fulfill the MIR density. This contour level is later on used in the procedure. The useful interval might be between 60 to 140% of the contour level. My advise is: try and see.

> set vari DENS_SCAL = 800.
> set vari MIN_STEP = 120
> set vari RE_PHASE = 15
> set vari RAD_SOLV = 4.5
> set vari RAD_COMB = 2.8
> set vari WEIGHT_FC = 1.1

The key "model_part" includes atoms model that will be refined and included into density combination procedure, while the key "model_mask" includes actually all the atoms places at the location where the model might be. Thereby we expand the model region and include a solvent flattening cycle too.

> key model_part select segm name DM_* end
> key model_mask select segm name D* end

> ener dens map MAP_FO dens scal DENS_SCAL
> set vari DONE = 0
> minimize sele work_set end step MIN_STEP writ 1 \
> phase steps RE_PHASE gain 90000 \
> phase macro dens_comb.cmds

> return

The MAIN_CMDS:dens_comb.cmds macro combines the MIR and partial model maps and calculated the difference map for the MINIMIZER. For description see the chapters "Density modification" section "Density combination".