muPPIt: De Novo Generation of Mutant-Specific Peptide Binders via Conditional Uniform Discrete Diffusion
Track: Full Paper Track
Keywords: Discrete diffusion, peptide design, Conditional sampling
TL;DR: muPPIt generates peptide binders specific to mutant proteins (vs. wild-type) from protein sequences alone via a transformer-guided uniform discrete diffusion model.
Abstract: The ability to selectively target disease-causing mutations in proteins, such as oncogenic mutations in cancer or pathogenic mutations in neurodegenerative diseases, is crucial for developing precise therapeutics that minimize off-target effects. Current approaches often lack the specificity required to distinguish between mutant and wildtype proteins, particularly in the absence of detailed structural information. In this work, we introduce muPPIt, a mutant-specific PPI targeting algorithm designed for the de novo generation of mutant-specific peptide binders based solely on mutant and wildtype sequences. At the core of muPPIt is MutBind, an attention-based model that differentiates between mutant and wildtype protein language model embeddings, achieving over 70% test accuracy in predicting binding probabilities. Additionally, we present PepUDLM, a uniform diffusion language model that generates diverse and biologically plausible peptides. By integrating MutBind’s predictions into PepUDLM’s sampling process, muPPIt efficiently designs peptides that specifically bind to mutant proteins. We demonstrate muPPIt’s effectiveness in computationally designing mutant-specific binders for a range of targets, including disease-related protein variants. In total, muPPIt serves as a powerful tool for developing highly specific peptide therapeutics, enabling precise targeting of mutant proteins without relying on structural information or structure-dependent latent spaces.
Attendance: Tong Chen
Submission Number: 28
Loading