Assessing sequence-based protein-protein interaction predictors for use in therapeutic peptide engineering

Area of Science:

• Computational biology
• Drug discovery
• Biotechnology

Background:

• Engineering peptides for targeted therapeutic effects is complex.
• Current in silico methods often lack target specificity.
• In silico screening offers a cost-effective alternative to experimental methods.

Purpose of the Study:

• To evaluate sequence-based protein-protein interaction (PPI) predictors for screening therapeutic peptides.
• To compare the efficacy of similarity-based versus deep learning methods in peptide design.
• To identify optimal strategies for in silico peptide engineering.

Main Methods:

• Curated a dataset of FDA-approved peptides.
• Assessed various sequence-based PPI predictors.
• Evaluated predictor performance in the context of peptide therapeutic engineering.
• Compared similarity-based approaches with deep learning models.

Main Results:

• Similarity-based PPI predictors are more suitable for screening therapeutic peptides than current deep learning methods.
• The effectiveness of this approach is highest when designing peptides for targets with known natural interactors.
• De novo peptide engineering may necessitate additional target-specific data.

Conclusions:

• Similarity-based PPI predictors are valuable tools for peptide therapeutic engineering, particularly for analog design.
• This study provides evidence supporting the use of these predictors in drug development pipelines.
• Future work may focus on enhancing de novo design capabilities with specialized datasets.