Targeted protein degradation of pathogenic proteins represents a powerful new treatment strategy for multiple disease indications. Unfortunately, a sizable portion of these proteins are considered “undruggable” by standard small molecule-based approaches, including PROTACs and molecular glues, largely due to their disordered nature, instability, and lack of binding site accessibility. As a more modular, genetically-encoded strategy, designing functional protein-based degraders to undruggable targets presents a unique opportunity for therapeutic intervention. In this work, we integrate pre-trained protein language models with recently-described joint encoder architectures to devise a unified, sequence-based framework to design target-selective peptide degraders without structural information. By leveraging known experimental binding protein sequences as scaffolds, we create a Structure-agnostic Language Transformer & Peptide Prioritization (SaLT&PepPr) module that efficiently selects peptides for downstream screening.
Please read and cite our manuscript published in Communications Biology!
All manuscript data/pipelines, as required/indicated in the manuscript Availability statements, are available on Zenodo and in the .zip file within this repo.
You can now access the model, training, and inference code on HuggingFace after signing the UbiquiTx License.