moPPIt-v3: Motif-Specific Peptides Generated via Multi-Objective-Guided Discrete Flow Matching

Precise targeting of therapeutic proteins to specific subsequence motifs within disease-related targets, such as conserved viral epitopes or mutant transcriptional domains, is critical for improving treatment efficacy and minimizing off-target interactions. Current computational binder design methods struggle to achieve this specificity, especially without reliable structural information. Here, we introduce moPPIt-v3, a generative, sequence-only model capable of the de novo design of high-affinity, motif-specific peptide binders. By coupling our Multi-Objective-Guided Discrete Flow Matching (MOG-DFM) framework with a residue-level interaction predictor, BindEvaluator, and a pretrained affinity predictor, we can guide peptide generation towards both sequence specificity and binding affinity. BindEvaluator is a transformer-based model, trained on over 510,000 annotated protein-protein interactions, that interpolates protein language model embeddings of two proteins via a series of multi-headed self-attention blocks, with a key focus on local motif features. BindEvaluator accurately predicts target binding sites given protein-protein sequence pairs with a test AUC > 0.94, improving to AUC > 0.96 when fine-tuned on peptide-protein pairs. By integrating BindEvaluator, we demonstrate moPPIt-v3's in silico efficacy by designing high-quality binders to specific motifs within target sequences with and without known peptide binders, including both structured and disordered targets. Moreover, we validate the motif-specificity of moPPIt-generated peptides in vitro by showing sensitive and specific binding toward distinct domains of cancer receptor NCAM1. Altogether, moPPIt-v3 is a powerful tool for developing highly-specific peptide therapeutics without relying on target structure or known binding partner.