Cutting protein targets helps design new binders
A new approach uses epitope excision and agent-guided optimization to design binders for large proteins.
Researchers used epitope excision to design binders for large proteins. This method works on 15 targets and 3,666 designs. It helps design binders against the 719-residue extracellular domain of fibroblast activation protein alpha (FAP).
Epitope excision truncates targets to local interface windows. This speeds up backbone diffusion and structure prediction. The excised windows closely reproduced full-target binding interfaces with a median RMSD of Å.
An agentic optimization loop helped the system autonomously identify catalytic-site targeting strategies. A human-in-the-loop (HITL) workflow raised the computational hit rate from to . This workflow also cut GPU hours by ~ .
Omitting target multiple sequence alignments during refolding increased the apparent yield. Wet-lab validation using yeast display identified nanomolar binders. The HITL and the fully autonomous agent (human-out-of-the-loop or HOOTL) recovered more enriched designs than the PI-alone baseline.
De novo protein design campaigns are bottlenecked by the inference compute time needed to generate and rank candidates. This limits wider functional exploration. The new method offers a way to scale de novo binder design to large, therapeutically relevant proteins.
Key facts
- Epitope excision cut RFdiffusion and AlphaFold3 GPU time by ~ in one of the epitopes.
- The HITL workflow raised the computational hit rate from to .
- Yeast display identified nanomolar binders from the generated designs.