SaLT&PepPr: An Interface-Predicting Language Model for Designing Peptide-Guided Protein Degraders

Protein-protein interactions (PPIs) are critical for biological processes and predicting the sites of these interactions is useful for both computational and experimental applications. We present a novel Structure-agnostic Language Transformer and Peptide Prioritization (SaLT&PepPr) pipeline to predict interaction interfaces from a protein sequence alone for subsequent generation of peptidic binding motifs. Our model fine-tunes the state-of-the-art ESM-2 protein language model (pLM) with a per-position prediction task to identify PPI sites using data from the PDB, and prioritizes motifs which are most likely to be involved within intra-chain binding. By only using amino acid sequence as input, the model is competitive with structural homology-based methods, but exhibits reduced performance compared with deep learning models that input both structural and sequence features, such as ScanNet. Inspired by our previous results using co-crystal structures to engineer target-binding "guide" peptide motifs derived from interaction sites on a binding partner, we use curated PPI databases to identify binding partners for subsequent peptide derivation. We specifically focus on several critical cancer-regulating proteins and prioritize guide peptides from either known or dataset-identified interacting partners. Fusing these guide peptides to an E3 ubiquitin ligase domain, CHIPΔTPR, we demonstrate degradation of endogenous β-catenin, 4E-BP2, and TRIM8, and specifically highlight the nanomolar binding affinity, low off-targeting propensity, and function-altering capability of our best performing degraders in cellular models of cancer. In total, our study suggests that prioritizing binders from natural interactions via pLMs can enable programmable protein targeting and modulation.