Target sequence-conditioned design of peptide binders using masked language modeling.

Area of Science:

• Computational biology
• Protein engineering
• Drug discovery

Background:

• Designing protein binders for 'undruggable' targets is challenging.
• Effective binder design often requires stable protein structures or structure-based models.

Purpose of the Study:

• Introduce PepMLM, a novel computational method for de novo linear peptide binder design.
• Enable the design of binders for any target protein without structural input.

Main Methods:

• PepMLM utilizes a target sequence-conditioned approach with a masking strategy.
• Fine-tuning the ESM-2 protein language model to reconstruct peptide binder regions.
• In silico benchmarking using AlphaFold-based docking and experimental validation through binding and degradation assays.

Main Results:

• PepMLM achieved low perplexities, comparable or superior to validated peptide-protein pairs.
• PepMLM-derived peptides showed sequence-specific binding to cancer and reproductive targets (e.g., NCAM1, AMHR2).
• Designed peptides enabled targeted protein degradation in models of Huntington's disease and viral infections.

Conclusions:

• PepMLM facilitates the design of de novo peptide binders for diverse therapeutic targets.
• The method's ability to bypass structural requirements broadens its applicability in drug development.
• PepMLM holds significant potential for advancing therapeutic development against previously inaccessible targets.