Page 174 - Macromolecular Crystallography
P. 174

MODEL BUILDING, REFINEMENT, AND VALIDATION  163

        the ‘rippling’ effect caused by Fourier transforms  by the ARP/wARP autobuilding module for con-
        with incomplete datasets, on top of depriving the  struction of a polypeptide chain that is in agreement
        optimizer algorithm of the refinement program from  with the electron density. In the regions where it
        valuable diffraction data.                   is not possible to make sense of the electron den-
                                                     sity features in terms of a protein model, free atoms
                                                     are kept. This results in what is called a hybrid
        11.2.5 Model building and refinement are
        both phase improvement procedures            model, a mixture of free atoms and fragments of
                                                     the protein structure. Much in the same way as in
        ARP/wARP challenged the common separation of  the ‘traditional’ X-ray structure solution, the hybrid
        model building and refinement by taking a more  model undergoes a refinement procedure with the
        general view of the underlying phase optimization  highly successful maximum likelihood based refine-
        through coordinate manipulation and by allowing  ment program REFMAC5 (Murshudov et al., 1997)
        the extension of the macromolecular model as part  in which the parameters of the model are adjusted
        of the whole process (Perrakis et al., 1999). This  to best fit the experimental data and stereochemical
        placed structure solution on a more unified foun-  expectations. If the quality of the model is suffi-
        dation and encouraged automation in linking the  ciently high, the phases improve overall and result
        entire procedure.                            in an enhanced electron density into which a more
          Model building is an interpretation of the cur-  accurate and more complete model may be built
        rently available electron density. Refinement is the  at the next cycle. ARP/wARP, like human crystal-
        adjustment of the built model to fit better to the  lographers, links model building and refinement
        experimental data. A crucial point here is that a  together into a unified process that iteratively pro-
        density map computed from the refined model is  ceeds towards a more complete macromolecular
        generally better than the map obtained from the  model.
        same model before the refinement. This then allows
        for an even better model to be built. Thus, refinement
        is needed to improve the outcome of model build-  11.3 Software packages
        ing by generating a better electron density map and  11.3.1 An overview of refinement programs
        model building is needed to provide a model in the
        first place and to provide stereochemical restraints  Refinement is an optimization problem. Therefore
        for the subsequent refinement to proceed smoothly.  we choose to outline the current programs based
        This viewpoint merges these two steps into one  on how the optimization problem is formulated
        model optimization process.                  (including the variables and the objective function)
          The ARP/wARP procedure is largely based on  and the method of choice for locating the min-
        the concept of describing the electron density with  ima (the solution method). The parameterization
        a collection of free atoms, originally introduced for  of a macromolecular model is a crucial step and
        phase extension by Isaacs and Agarwal (1985). The  should reflect the amount of available data that
        main a priori knowledge in this case is the fact that  can be used to optimize the model parameters.
        proteins consist of atoms, although this informa-  This fluid parameterization is typically replaced by
        tion alone is hardly sufficient.ARP/wARP, however,  standard atomic coordinates (xyz), a displacement
        takes this further – before linking the atoms, one  parameter (B), and occupancy (occ) parameteriza-
        has to position them correctly, not necessarily into  tion in which the amount of data is accounted
        peaks of the density but preserving at least the  for in the weights (tightness) of the stereochemical
        first order stereochemical information – the inter-  restraints. Other parameterization approaches use
        atomic distances. Free atoms themselves account  torsion angle, rigid body, and NCS constrained/
        for the electron density features without the need  restrained (hard/soft) parameterization. Additional
        to apply additional chemical knowledge or distance  parameters include those used for modelling the
        restraints between them. The coordinates of the free  bulk solvent. Within the free atoms framework, the
        atoms are refined, updated where needed, and used  atom type itself enters as a parameter since such
   169   170   171   172   173   174   175   176   177   178   179