Computationally Evidence-Grounded Sequence-First Design of Peptide Binders

Area of Science:

• Computational biology
• Protein engineering
• Drug discovery

Background:

• Peptide binders are crucial for targeting proteins intractable to small molecules.
• Current peptide discovery methods are limited by screening intensity or template dependence.
• Sequence-based generation offers scalability but often lacks target specificity and struggles with protein language model transfer.

Purpose of the Study:

• To introduce BOND-PEP, a sequence-only framework for generating peptide binders with enhanced target specificity.
• To address limitations of existing sequence-based peptide design methods.
• To enable controllable peptide binder generation without structural templates.

Main Methods:

• BOND-PEP employs retrieval of target-relevant peptide exemplars.
• It utilizes bipartite message passing for protein-peptide alignment.
• A protein-centric representation guides conditional decoding for binder generation.

Main Results:

• BOND-PEP demonstrated superior performance over RFdiffusion, PepPrCLIP, and PepMLM in AlphaFold-Multimer evaluations.
• The framework achieved improved reference-beating ipTM success on a held-out benchmark.
• Successful transfer to external targets with experimentally validated binders was observed.

Conclusions:

• BOND-PEP represents an evidence-grounded framework for sequence-only peptide binder generation.
• Retrieval-augmented, topology-conditioned decoding is a viable strategy for controllable peptide design.
• This approach offers a scalable and effective alternative for discovering peptide binders.