Target sequence-conditioned design of peptide binders using masked language modeling
Problem
Designing de novo linear peptide binders for disease targets, including 'undruggable' ones, is difficult without stable structural information.
Why it matters
Peptide binders could enable new therapies and targeted protein degradation across cancer and reproductive diseases.
What the AI did
PepMLM finetunes the ESM-2 protein language model with a C-terminal masking strategy to reconstruct binder sequences conditioned on target sequences; candidates were benchmarked with AlphaFold-based docking and tested experimentally.
Results
Low perplexities matching or improving upon validated peptide-protein pairs; sequence-specific binding to NCAM1 and AMHR2 and targeted protein degradation demonstrated in assays; numeric metrics not stated in source
Impact
AI-designed peptides showed experimentally validated binding and degradation activity against named disease targets.
Record
| Venue | Nature Biotechnology |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Tianlai Chen, Zachary Quinn, Madeleine Dumas, Christina Yi Peng, Lauren Hong, Moises Lopez-Gonzalez, Alexander A. Mestre, Rio Watson et al. |
| Citations | 59 (OpenAlex, at import time) |
| References | 57 |
| Model or method | PepMLM (finetuned ESM-2) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41587-025-02761-2 |
| Open access | hybrid |
| Keywords | peptide design, protein language model, ESM-2, masked language modeling, targeted degradation, AlphaFold docking |
Importance rating
significant — thousands to millions affected
De novo binders to 'undruggable' targets enable therapeutics and targeted degradation across cancer, reproductive, Huntington's, and viral contexts; broad therapeutic potential but no scale stated.