Enhancing cross-domain protein and peptide interaction with retrained deep learning models.

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Protein-peptide interactions (PPepIs) are crucial for biological processes and peptide therapeutics.
• Existing deep learning models for protein-protein interaction (PPI) prediction face challenges with generalization and overfitting.

Purpose of the Study:

• To develop a more accurate and efficient deep learning framework for predicting protein-peptide interactions (PPepIs).
• To address overfitting and generalization issues in deep learning-based interaction prediction.

Main Methods:

• Integrated protein sequence and structure information into a multilevel deep learning framework.
• Focused model training on short proteins with reduced sequence redundancy.
• Utilized experimentally validated protein-protein interaction (PPI) pairs from the STRING database to construct the training dataset.

Main Results:

• Training on short-protein datasets significantly improved prediction accuracy and computational efficiency compared to long-protein datasets.
• The retrained model successfully delineated human protein and SARS-CoV-2 virus PPI networks.
• Screening of drug peptides revealed numerous potential therapeutic targets and side effects.

Conclusions:

• The developed deep learning model offers a robust method for delineating PPepI networks.
• The model aids in identifying peptide drug targets, analyzing side effects, and investigating viral infections.
• This approach provides valuable resources for both therapeutic development and virology research.