Sequence (250 AA)
VQLVQSGAEVKKRGSSVKVSCKSDTNTNPLGAINWVRQAPGQGLEWMGGVRMNGGIANYAQKFQGRVTITTDESTSTAYMELSSLRSEDTAVFYGLDLNHSGEQLAPHPSQYYYYYYGMDVWGQGTTVTVSSGGGGSGGGGSGGGGSEIVMTQSPGTPSLSPGERATLSFTAGSGVTGDYLAWYQQKPGQAPRLLVTNAGTRATGIPDRFSGSGSGTDFTLTISRLEPEDFAVYYAALLKGGSSFGQGTK
No experimental data
This protein hasn't been validated in the lab yet.
This protein was designed using KPRFdiffusion–MPNN–Boltz–ipSAE
Template & antigen preparation Template complex: PDB 6CMI, using complete variable heavy and light chains of the antibody; non-essential heteroatoms and distant chains removed. Structural alignment: Hendra RBP (from 6CMI) structurally aligned to Nipah RBP; conserved / structurally similar interface residues on the viral RBP identified as key contact positions for optimization. Complex modeling: Nipah RBP + 6CMI-based antibody complex predicted with AlphaFold3, used as the input structure for all downstream design steps.
Design region and constraints Design region: Heavy-chain CDRs (CDRHs) defined by Chotia numbering;
Constraints:
Framework residues and non-target CDRs fixed to 6CMI sequence.
CDR lengths kept identical to 6CMI (length-preserving design).
Antigen (Nipah RBP) backbone and sequence kept fixed during design.
RFdiffusion
Mode: local inpainting / binder design on CDRH3 only (mask = CDRH3; framework and antigen frozen).
Objective: generate backbone variants that improve packing against the conserved Nipah RBP interface while preserving the global binding geometry inherited from 6CMI.
Hyperparameters: RFdiffusion default binder-design settings, applied to the masked CDRHs region with length preserved.
ProteinMPNN
Backbone: RFdiffusion-designed antibody–antigen complexes.
Mutability: only CDRHs residues allowed to change; all other antibody residues fixed to 6CMI sequence.
Goal: generate sequence variants compatible with the redesigned CDRH3 backbone while maintaining the original scaffold.
Boltz-2 structure prediction
For each designed sequence, we predict antibody–Nipah RBP complexes with Boltz-2 under multiple settings to capture structural uncertainty:
Runs with ~200 recycles Runs with ~500 recycles
For each configuration, multiple models / seeds are generated to build a small ensemble per sequence.
ipSAE and ranking
For every Boltz-2 model, we compute ipSAE and related interface metrics. We observe large variability in ipSAE across models for the same sequence (values ranging roughly from ~0.1 to ~0.7), so we: Record ipSAE per model and ipSAE_min / ipSAE_best per sequence. Rank designs using ensemble statistics (e.g., best ipSAE and consistency across models), not a single structure.
Final candidates submitted are those with favorable ipSAE, good structural consistency across Boltz-2 runs (200 vs 500 recycles)