Atomically accurate de novo design of antibodies with RFdiffusion (www.nature.com)

🤖 AI Summary
Researchers have demonstrated the first fully de novo, epitope-specific antibody design pipeline by fine-tuning RFdiffusion to generate antibody variable domains (VHHs), scFvs and full antibodies that bind user-specified epitopes with atomic-level precision. Designs to multiple disease-relevant sites (influenza haemagglutinin, C. difficile TcdB, RSV, SARS-CoV-2 RBD and a PHOX2B peptide–MHC) were experimentally validated by yeast-display, SPR and cryo‑EM; cryo‑EM and a high‑resolution structure confirmed the designed CDR conformations and binding poses. Although initial computational designs typically showed modest affinities (tens–hundreds of nM), in vitro affinity maturation using OrthoRep produced single‑digit nanomolar binders that preserved intended epitope selectivity. This establishes a path from in silico design to therapeutic-grade binders without requiring a starting antibody or immunization. Technically, the team fine‑tuned RFdiffusion on antibody–antigen complexes while conditioning on a fixed framework provided via RF2’s template track (pairwise distances/dihedrals) and a hotspot one‑hot to specify epitopes; RFdiffusion samples CDR loop conformations and antibody rigid‑body placement using the AlphaFold/RF frame representation and the diffusion noising/denoising schedule. ProteinMPNN was used to sequence CDRs post‑design, and a fine‑tuned RoseTTAFold2 (RF2) served as a powerful filter by predicting holo antibody–antigen structures for self‑consistency. Key implications: de novo epitope targeting and atomic‑precision CDR design can accelerate therapeutic discovery and enable targeting of epitopes difficult to obtain via traditional library or immunization methods, though the approach currently depends on holo target structures, experimental screening, and subsequent affinity maturation to reach clinical affinities.
Loading comments...
loading comments...