UMPPI: Unveiling Multilevel Protein-Peptide Interaction Prediction via Language Models

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Protein-peptide interactions are vital for cellular functions and disease pathogenesis.
• Accurate identification of binding residues is crucial for peptide function analysis and therapeutic development.
• Current computational methods face limitations in feature integration and data dependency.

Purpose of the Study:

• To introduce UMPPI (Unveiling Multilevel Protein-Peptide Interaction), a novel framework for simultaneous prediction of protein-peptide interactions and binding residues.
• To leverage the pretrained protein language model ESM2 for enhanced sequence and structural information encoding.
• To overcome the limitations of existing methods by integrating sequence and structure-based approaches.

Main Methods:

• Developed a multiobjective optimization framework, UMPPI.
• Utilized the ESM2 protein language model for sequence encoding and latent structural feature extraction.
• Implemented a strategy for simultaneous prediction of binary interactions and binding residues.

Main Results:

• UMPPI accurately predicted binary protein-peptide interactions.
• UMPPI successfully identified critical binding residues on both protein and peptide partners.
• The framework effectively integrated sequence information with latent structural insights.

Conclusions:

• UMPPI offers a robust computational tool for protein-peptide interaction prediction.
• The identification of binding residues by UMPPI facilitates a deeper understanding of peptide function.
• UMPPI shows significant potential in advancing the peptide drug discovery pipeline.