moPPIt: De Novo Generation of Motif-Specific Binders with Protein Language Models

Abstract

Targeting specific functional motifs, whether conserved viral epitopes, intrinsically disordered regions (IDRs), or fusion breakpoints, is essential for modulating protein function and protein-protein interactions (PPIs). Current design methods, however, depend on stable tertiary structures, limiting their utility for disordered or dynamic targets. Here, we present a motif-specific PPI targeting algorithm (moPPIt), a framework for the de novo generation of motif-specific peptide binders derived solely from target sequence data. The core of this approach is BindEvaluator, a transformer architecture that interpolates protein language model embeddings to predict peptide-protein binding site interactions with high accuracy (AUC = 0.97). We integrate this predictor into a novel Multi-Objective-Guided Discrete Flow Matching (MOG-DFM) framework, which steers generative trajectories toward peptides that simultaneously maximize binding affinity and motif specificity. After comprehensive in silico validation of binding and motif-specific targeting, we validate moPPIt in vitro by generating binders that strictly discriminate between the FN3 and IgG domains of NCAM1, confirming domain-level specificity, and further demonstrate precise targeting of IDRs by generating binders to the N-terminal disordered domain of {beta}-catenin. In functional, disease-relevant assays, moPPIt-designed peptides specifically targeting the GM-CSF receptor [a] subunit effectively block human macrophage polarization. Finally, we demonstrate utility in cell engineering, where binders directed against a defined motif on a synthetic cell surface ligand (AGR2t) drive specific chimeric antigen receptor regulatory T cell (CAR Treg) activation and suppressive function, including in the context of human induced pluripotent stem cell (hiPSC)-derived cardiomyocytes. Altogether, moPPIt serves as a theoretically-justified, sequence-based paradigm for controllably targeting the complete proteome with immediate therapeutic applications.