Fig. 3: Experimental characterization of multiepitope immunogen candidates.
From: Accurate single-domain scaffolding of three nonoverlapping protein epitopes using deep learning
a, In silico evaluation. AF2 is used to assess the similarity of predicted structures to the design model and the similarity of each epitope within the prediction to that of the native epitope. Vertical lines: threshold for in silico ‘success’. b, Representative design generated with RFjoint2 showing the epitopes remaining accessible to their target antibodies despite this not being explicitly specified during design. The predicted structure of RSVF-multi-4 is aligned to the three known epitope-targeting antibodies (site II, PDB 3IXT; site IV, PDB 3O41; site V, 5TPN). c, Predicted structures of top candidate multimotif scaffolds. d, CD spectra at start and end incubation temperatures shown for each scaffold. e–g, Normalized SPR steady-state affinity measurements for four scaffolds for each epitope-specific antibody: RSV90 (site V), motavizumab (site II) and 101F (site IV).
