Accurate de novo design of high-affinity protein-binding macrocycles using deep learning

Area of Science:

• Computational biology
• Protein engineering
• Drug discovery

Background:

• Developing macrocyclic binders for therapeutic proteins often requires resource-intensive screening with limited control over binding modes.
• Current de novo design approaches for protein-binding macrocycles lack robustness.

Purpose of the Study:

• To introduce RFpeptides, a denoising diffusion-based pipeline for the de novo design of macrocyclic binders against protein targets.
• To demonstrate the efficacy of RFpeptides in designing high-affinity binders for therapeutic applications.

Main Methods:

• Utilized a denoising diffusion model (RFpeptides) for computational design of macrocyclic peptides.
• Tested designed macrocycles against four diverse protein targets, including Rhombotarget A (RbtA).
• Validated binding affinity and structural accuracy using experimental methods, including X-ray crystallography.

Main Results:

• Successfully obtained macrocyclic binders with medium to high affinity against all four tested protein targets.
• Designed a high-affinity binder (K_d < 10 nM) for RbtA from a predicted structure.
• Experimental structures of macrocycle-protein complexes closely matched computational models (Cα RMSD < 1.5 Å).

Conclusions:

• RFpeptides offers a robust framework for the rapid, custom design of macrocyclic peptide binders.
• The pipeline facilitates the development of novel macrocyclic peptides for diagnostic and therapeutic applications.
• This approach overcomes limitations of traditional screening methods in protein-binding macrocycle development.