Visual automated macromolecular model building

Abstract

Automated model-building software aims at the objective interpretation of crystallographic diffraction data by means of the construction or completion of macromolecular models. Automated methods have rapidly gained in popularity as they are easy to use and generate reproducible and consistent results. However, the process of model building has become increasingly hidden and the user is often left to decide on how to proceed further with little feedback on what has preceded the output of the built model. Here, ArpNavigator, a molecular viewer tightly integrated into the ARP/wARP automated model-building package, is presented that directly controls model building and displays the evolving output in real time in order to make the procedure transparent to the user.

Keywords: ARP/wARP; model building; molecular graphics.

Figure 1

Workflow of ArpNavigator building a model of the 475-residue protein leishmanolysin (PDB entry 1lml; Schlagenhauf et al., 1998 ▶) using 2 Å resolution data. (a) The free-atoms model is placed into the experimental density map and is used for iterative chain tracing to build the protein backbone, resulting in a hybrid model (b). The refinement and rebuilding of the hybrid model are repeated iteratively, resulting in a more complete model (c). All coordinate files and maps generated during this process are loaded in real time into ArpNavigator.

Figure 2

The log window that displays detailed information while building the protein structure shown in Fig. 1 ▶.

Figure 3

A selection of model-viewing options in ArpNavigator depicted for PDB structure 3l9a: the C-terminal domain of a Streptococcus mutans hypothetical protein (Midwest Center for Structural Genomics, unpublished work). Shown clockwise from the top left are a stick representation in solid electron density (50% opacity, 1.5σ), a ball-and-stick representation in planar density, also showing a ‘scaleometer’ and the contour levels of shown electron densities, a skeleton representation of the electron density shown as a mesh and the protein in cartoon representation in planar density.

Figure 4

Helices built using the right-click ‘quick action’ menu. This helical substructure, computed in a few seconds on a single desktop workstation, was built into a 3.0 Å resolution electron-density map calculated from the experimental data associated with PDB entry 1c48 (345 residues; mutated shiga-like toxin; Ling et al., 1998 ▶).

Figure 5

The modelling of the ligand flavin adenine dinucleotide into p-hydroxybenzoate hydroxylase (PDB entry 1cc6; Eppink et al., 1999 ▶) as depicted by the ArpNavigator GUI. (a) Following input of the experimental 2.2 Å resolution data and the protein model, a difference density map is automatically calculated and shown along with the ‘sparse grid’ used for ligand building by the ARP/wARP ‘label-swapping’ method. (b) The ensemble of models built prior to selection of the best-fitting ligand model. (c) The final ligand model after real-space refinement is output to the screen for viewing and further analysis where necessary.

Figure 6

Integrative model building with ArpNavigator. Assessing the quality of initial electron density by building helices (a) or skeletons (b). Following model building, in the case of bound ligands the binding site can be identified (c) and the coarsely placed ligand refined into the map (d). Solvent molecules can be modelled (e). Contour levels are at 1.5σ in (a) and (b) and at 3.5σ in (c) and (d). The resolution of the maps used is 3.2 Å in (a) and 1.9 Å in (b), (c) and (d); all maps shown are 2mF _o − DF _c maps.

Figure 7

Comparison of 2mF _o − DF _c electron-density maps contoured at 1.5σ above the mean and models of an alanine-glyoxylate aminotransferase (AGT; Fodor et al., 2012 ▶) built at 2.0 Å resolution (PDB entry 3r9a) in the correct space group (a) and the wrong space group (b).