Page 176 - Macromolecular Crystallography
P. 176

MODEL BUILDING, REFINEMENT, AND VALIDATION  165

        was one of the first successful approaches to auto-  thenbuiltinthesesecondarystructureregionswhilst
        mate construction of new protein structures from  taking care not to over-interpret them. Secondary
        previously seen fragments, following the syntac-  structural main-chain fragments of sufficient length
        tic concept of pattern recognition. FFFear (Cowtan,  (15–20 residues) are slid along the sequence and the
        1998) presented an elegant implementation of the  best match to the side-chain density is used to assign
        ESSENS idea in a fast Fourier framework that  theresiduetype.Loopsarebuiltbyextendingthebuilt
        allowed not only to speed-up program execution but  main-chain fragments by sampling Ramachandran
        also the use of structure templates in a probabilistic  space.
        manner.                                       The ARP/wARP package follows the full path of
          The software packages ARP/wARP (Lamzin and  syntactic construction of a model starting from free
        Wilson, 1993; Perrakis et al., 1997, 1999, 2001; Lamzin  atoms, through candidate Cα atoms, peptide and
        et al., 2001; Morris et al., 2002, 2003, 2004; Cohen et al.,  dipeptide units, chain fragments to sequence dock-
        2004, Zwart et al., 2004), TEXTAL/CAPRA (Ioerger  ing, and side-chain building. It was the first software
        etal., 1999; Holtonetal., 2000; IoergerandSacchettini,  package to combine most steps of model building
        2002, 2003), MAID (Levitt, 2001), and RESOLVE  with refinement in an iterative manner. The power
        (Terwilliger, 2001, 2003a, 2003b, 2003c, 2004) aim to  of ARP/wARP lies precisely in this iterative nature
        deliver a complete macromolecular model starting  (Perrakis et al., 1999). Although iteration is offered
        from an electron density map. RESOLVE is perhaps  in other packages it is de facto less important and
        one of the most advanced programs in terms of  less powerful due to the lack of the use of the free
        implementingaproperstatisticalpattern-recognition  atoms concept and the focus on a ‘snapshot’ inter-
        system. A promising development of RESOLVE   pretation of the current density map. The concept
        involves the use of template matching of a number  of ARP/wARP is different. When atoms are placed
        of commonly observed density distributions to per-  in electron density only the fragments that could be
        form density modification. FFT-based methods are  interpreted with confidence are built and refined as
        employed to calculate derivatives of a log-likelihood  part of the restrained atomic model. The parts that
        map,andstructurefactorstatisticsarecombinedwith  were less certain are still allowed to move as free
        prior knowledge of density distributions to provide  atoms. Atoms that do not refine well can be removed
        scores, which are used to match templates to grid  and density that is yet unaccounted for can be filled
        points and their surrounding region to perform den-  with atoms at any point. In this way, not only the
        sity modification. TEXTAL uses a high-dimensional  atomic parameters but the atoms themselves and the
        feature space consisting of rotationally invariant  connectivity between them are made part of the opti-
        properties derived from the electron density. These  mization that so encompasses both model building
        includetheaveragedensityandhighermoments, the  and refinement.
        varianceovergridpoints, themomentsofinertiaand  For a detailed comparison of some popular
        spokes of density values (the three bonding tubes of  automated model building packages, see Badger
        density that are to be expected around a Cα atom).  (2003). For an analysis of structural features and
        TEXTALcomputesthesefeaturevectorsandattempts  their relationship to model building strategies, see
        to recognize the closest density pattern from a large  Morris (2004). Further details should be sought in
        database of precomputed patterns. After the density  the original articles.
        featureshavebeenmatched, thecoordinatesfromthe
        database template are used to place atoms into den-
                                                     11.3.3 Automation in crystallographic structure
        sity. The program MAID relies on skeletonization of
                                                     solution: decision-making systems
        agivenelectrondensitymaptodeterminethepathof
        the main chain. The skeleton grid points then serve  An emerging goal in the field of macromolecular
        as features in which secondary structural elements  crystallography methods development is the overall
        patterns are sought, very much like the computer  automation of the time-consuming steps in structure
        graphics model building steps with a program such  solution. To a large extent, two software packages
        as O. A stereochemically accurate atomic model is  exemplified that concept at the turn of the previous
   171   172   173   174   175   176   177   178   179   180   181