6 Nov 2025

Fine-tuned RFdiffusion designs antibody fragments that bind chosen epitopes

Biotechnology and synthetic biologyExperimental cross-disciplinary systems
AI illustration · Illustrative scene, not a photograph of the specific device or experiment.

What was demonstrated

A fine-tuned RFdiffusion network, paired with yeast display screening, generated VHH antibody fragments and single-chain variable fragments (scFvs) designed to bind user-specified epitopes, according to a bioRxiv preprint. VHHs were characterised against four disease-relevant epitopes. Cryo-electron microscopy confirmed the designed binding poses for VHHs against influenza haemagglutinin and Clostridium difficile toxin B (TcdB), and a high-resolution structure of the influenza-binding VHH matched the designed complementarity-determining regions. Initial designs showed modest affinity, and the abstract gives no dissociation constants. Affinity maturation with OrthoRep produced single-digit nanomolar binders that kept epitope selectivity. Designed scFvs against TcdB and a PHOX2B peptide-MHC complex were also reported, with cryo-EM confirming binding poses for two TcdB scFvs.

Why it matters

The preprint suggests that computational design followed by screening can produce antibody fragments against a chosen site. Whether success rates can be raised enough to matter in practice is not shown here. The work does not show that de novo designed antibodies can replace conventional antibodies in therapeutic use.

Caveats

  • The main source is a bioRxiv preprint. The version checked had not been peer reviewed, and the journal version could not be opened during checking.
  • Initial designs bound with modest affinity. Single-digit nanomolar binders needed laboratory affinity maturation, so the design step alone did not produce binders at that affinity.
  • Phys.org, a news report, gives a success rate range of 0% to 2% across targets rather than per-target figures. It also reports that one SARS-CoV-2 design bound at a different angle from the one planned, which the authors labelled a design failure.
  • The structural work covers a small number of designs and four epitopes. The preprint does not report clinical testing.
  • Phys.org reports that a designed binder against a cancer-related peptide-MHC target was recognised in two assays, but engineered T cells built from those designs did not kill tumour cell lines in laboratory tests.

Technology profiles

Sources

  1. preprint
    Atomically accurate de novo design of antibodies with RFdiffusion
    Nathaniel R. Bennett, Joseph L. Watson, Robert J. Ragotte et al.. bioRxiv, 28 Feb 2025. DOI: 10.1101/2024.03.14.585103.
  2. news
    AI-designed antibodies created from scratch
    Justin Jackson. Phys.org, 6 Nov 2025.

Last reviewed 21 Oct 2026.