Refinement at high resolution (higher than approx. 1.0 Angstrom)

Guidelines for structure refinement at high resolution:

• make sure the model contains hydrogen atoms. If not, phenix.reduce can be used to add them:

  % phenix.reduce model.pdb > model_h.pdb
  

  By default, phenix.refine will refine positions of H atoms as riding model (H atom will exactly follow the atom it is attached to). Note that phenix.refine can also refine individual coordinates of H atoms (can be used for small molecules at ultra-high resolutions or for refinement against neutron data). This is governed by hydrogens.refine_sites = individual *riding keyword and the default is to use riding model. hydrogens.refine_adp defines how hydrogens' B-factors are refined (default is to refine one group B for all H atoms). At high resolution one should definitely try to use one_b_per_molecule or even individual choice (resolution permitting). Similar strategy should be used for refinement of H's occupancies, hydrogens.refine_occupancies keyword.

• most of the atoms should be refined with anisotropic ADP. Exceptions could be model parts with high B-factors), atoms in alternative conformations, hydrogens and solvent molecules. However, at resolutions higher than 1.0A it's worth of trying to refine solvent with anisotropic ADP.

• it is a good idea to constantly monitor the existing solvent molecules and check for new ones by using ordered_solvent=true keyword. If it's decided to refine waters with anisotropic ADP then make sure that the newly added ones are also anisotropic; use ordered_solvent.new_solvent=anisotropic (default is isotropic). One can also ask phenix.refine to refine occupancies of water: ordered_solvent.refine_occupancies=true (default is False).

• at high resolution the alternative conformations can be visible for more than 20% of residues. phenix.refine automatically recognizes atoms in alternative conformations (based on PDB records) and by default does constrained refinement of occupancies for these atoms. Please note, that phenix.refine does not build or create the fragments in alternative conformations; the atoms in alternative conformations should be properly defined in input PDB file (using conformer identifiers) (if actually found in a structure).

• the default weights for stereochemical and ADP restraints are most likely too tight at this resolution, so most likely the corresponding values need to be relaxed. Use wxc_scale and wxu_scale for this; lower values, like 1/2, 1/3, 1/4, ... etc of the default ones should be tried. phenix.refine allows automatically optimize these values ( optimize_wxc=True and optimize_wxu=True), however this is a very slow task so it may be considered for an over night run or even longer. At ultra-high resolutions (approx. 0.8A or higher) a complete unrestrained refinement should be definitely tried out for well ordered parts of the model (single conformations, low B-factors).

• at ultra-high resolution the residual maps show the electron density redistribution due to bonds formation as density peaks at interatomic bonds. phenix.refine has specific tools to model this density called IAS models (Afonine et al, Acta Cryst. (2007). D63, 1194-1197).

This example illustrates most of the above points:

% phenix.refine model_h.pdb data.mtz high_res.params

where the file high_res.params contains following lines (for more parameters under each scope look at complete list of parameters):

refinement.main {
  number_of_macro_cycles = 5
  ordered_solvent=true
}
refinement.refine {
  adp {
    individual {
      isotropic = element H
      anisotropic = not element H
    }
  }
}
refinement.target_weights {
  wxc_scale = 0.25
  wxu_scale = 0.3
}
refinement.hydrogens {
  refine_sites = individual *riding
  refine_adp = *one_b_per_residue  one_b_per_molecule  individual
  refine_occupancies = *one_q_per_residue  one_q_per_molecule  individual
}
refinement {
  ordered_solvent {
    mode = auto filter_only *every_macro_cycle
    new_solvent =  isotropic *anisotropic
    refine_occupancies = True
  }
}

In the example above phenix.refine will perform 5 macro-cycles with ordered solvent update (add/remove) every macro-cycles, all atoms including newly added water will be refined with anisotropic B-factors (except hydrogens), riding model will be used for positional refinement of H atoms, one occupancy and isotropic B-factor will be refined per all hydrogens within a residue, occupancies of waters will be refined as well, the default stereochemistry and ADP restraints weights are scaled down by the factors of 0.25 and 0.3 respectively. If starting model is far enough from the "final" one, more macro-cycles may be required (than 5 used in this example).