Fig 1.
General software structure of RPXDock.
A. User-defined inputs are given as options to the dock.py application. B. Within the application, input .pdb files are stored in the Body object as a PyRosetta pose. The Body class implements a Bounding Volume Hierarchy (BVH) for rapid operations on coordinates. C. The Spec and Sampler classes define the rigid-body DOFs the Body object is allowed to sample. D. Within the Search class, the Body object receives the DOFs as rigid body transforms (indicated as grid squares). Each transform is evaluated by the Motif and Score classes, which ranks the quality of residue-pair motifs at a given interface of a dock [17] and subsequently summarizes the residue-pair motif scores with additional interface quality metrics through a user-selected score function. The top scoring transforms are searched iteratively with higher resolution sampling and scoring in a hierarchical search algorithm. E. The final top scoring transforms from the search are fed into the Result class, which prunes the results using filter metrics and clusters the transforms based on backbone redundancy. F. The results are stored and output as transforms, which can be re-applied to the input Body object to generate a full-atom .pdb file of the resulting docked configuration.
Table 1.
Keywords associated with each currently supported architecture.
Fig 2.
Example inputs and docking output architectures currently supported by RPXDock.
X/Y/Z cartesian axes are shown in red, green, and blue respectively. Corresponding translational and rotational DOFs are sampled along and around these axes. Axes where DOFs are not sampled for an architecture are colored gray. A. Asymmetric docking samples 6 DOFs belonging to the first of two input monomers. B. Cyclic docking samples four DOFs belonging to an input monomer to generate a cyclic structure with its cyclic axis aligned to the Z axis. C-F. Oligomeric input structures must have their cyclic axis aligned to the Z axis and the input .pdb should only contain the asu (dark). Stacking, dihedral, polyhedral group, and wallpaper docking samples the rotational and translational DOF along the Z axis of the input cyclic oligomer, which is aligned during docking to the relevant rotational symmetry axes in the target architecture.
Fig 3.
Schematic representation of hierarchical sampling.
A. Schematic of a search grid for a single DOF keeping only the transforms that passed hierarchical scoring (blue) at each stage of search. This reduces the space searched at later stages where the search grid is subdivided at increasing resolution. B. A schematic depiction of protein backbones sampled with increasing resolution. The backbones shown would correspond to a single blue box at each stage of the search depicted in panel A; a cloud of such backbones would be sampled for each of the distinct docked configurations corresponding to each blue box. C. Residue-pair motifs are also evaluated at increasing resolution during each iteration of the search.
Fig 4.
Interface size bias by the sasa_priority score function.
A. 572 pairs of inputs were docked in a two-component icosahedral architecture at a–-weight_sasa value of 900, 1200, 1500, 1800, 2100, and 2400, with total area under each curve normalized to 1. B. The interface of the top-scoring docked configuration for ‐‐weight_sasa value of 900, 1500, and 2100 is highlighted (green). Estimated buried SASA calculated using Rosetta for these docks are 864, 1416, and 1795 Å2, respectively.
Table 2.
List of additional score functions.
Fig 5.
Docking and characterization of one- and two-component polyhedral assemblies using RPXDock.
A. Models of one- and two-component docked polyhedral assemblies with the oligomeric building blocks in purple and orange. The asymmetric unit of each assembly, comprising one subunit of each building block, is colored dark purple and dark orange. B. Reference-free 2D class averages from negative stain electron microscopy. Each assembly is viewed along several axes of symmetry. C. 3D density maps reconstructed from selected 2D class averages. D. Overlays of each design model fit into its 3D density map, confirming that each design assembles to the architecture identified by RPXDock. E. Characterization of the two-component octahedral assembly O43-rpxdock-EK1 by cryogenic electron microscopy. The design model is colored as in A). To the right are representative 2D class averages showing different axes of symmetry and a reconstructed 3D map at 3.7 Å resolution. The overlay of the original dock (orange and purple) with the model built from the 3D reconstruction (gray) shows 4 Å Cα root mean square deviation between the original dock and cryoEM structure over 48 chains.