Fig 1.
Computation of the surface similarity score (SurfS).
Protein surface is converted into a point cloud where each point is used to compute shape and electrostatic features. To compute the SurfS score, all individual points of the point clouds are compared and the shape similarity value is derived from closest points of the two surfaces in space while electrostatic similarity is evaluated by correlation analysis of the electrostatic potentials of both surfaces.
Fig 2.
Single mutant discrimination using surface similarity score in protein-protein complexes.
A) Surface similarity evaluation protocol for single amino acids. B) Recovery for all 19 considered amino acid types in bound (top) and unbound (bottom) complex states, evaluated with four different metrics: SS (shape similarity), REU (Rosetta energy unit), ES (electrostatic similarity), and SurfS (surface similarity). C) Average surface similarity score when performing all-against-all amino acid comparison for bound (top) and unbound (bottom) complex states. The highest mean SurfS score for every amino acid is highlighted.
Fig 3.
Sequence recovery of protein interfaces.
A) Sequence recovery benchmark pipeline. Sequences in the interfaces of protein-protein complexes are evaluated in the presence and absence of the binding partner. The tested complexes were grouped into interfaces with low and high shape complementarity, and antibody-antigen complexes. Surface-centric design (RosettaSurf and RosettaSurf-site) is compared to a standard structural protein design protocol (FixBB). B) Interface sequence recovery of the complete dataset. C) Sequence recovery of low shape complementarity interfaces. D) Sequence recovery of high shape complementarity interfaces. E) Sequence recovery of antigen-antibody complexes. Dashed lines represent median and triangles represent mean recovery values.
Fig 4.
Comparison of SSM data obtained for the designed interleukin-2/15 antagonists in comparison to RosettaSurf-site predictions.
The structure highlights the five selected positions of the interleukin design that were computationally and experimentally sampled. Different sampling results of the experimental SSM, RosettaSurf-site with SC, and Rosetta’s energy function are reported in the table. Mutations resulting in the experimentally reported best binding design are highlighted in dark green. RosettaSurf-site was able to recover four out of five key binding mutations (dark green) while evaluating mutations with Rosetta’s energy function could only retrieve one binding mutation at position 98. Additionally, RosettaSurf-site was able to recover four affinity improving mutations not present in the best binding design, whereas Rosetta’s energy function could identify only two of these mutations (light green).
Fig 5.
Surface centric design of a viral antigenic site present the in RSVF.
A) Design process of site 0-mimicking protein scaffolds. Starting scaffolds are selected from the PDB based on structural alignments with the epitope helix. The surface mimicking designs are generated by grafting the side chains of the helix segment of the epitope onto the scaffold and surface-centric design is employed to optimize the loop region. Before and after design of the surface compared to native site 0. Blue areas indicate high similarity. B) Mimicry of surface geometry of WT scaffold, RSV_FixBB, and Surf_03 designs compared to native site 0. C) Representative SPR measurements of Surf_03 and RSV_FixBB against site 0-specific antibodies D25 and ADI14496. D) Binding profiles of Surf_03, a FixBB designed protein, and a helix-only design against a panel of site-specific antibodies with green indicating binding and red cells corresponding to non-binding. A knockout mutant of Surf_03 and the WT protein binding profiles are listed as reference.