Cat-PIPpred: Pro-Inflammatory Peptide Predictor Integrating CatBoost and Cross-Modal Feature Fusion.

Area of Science:

• Biochemistry
• Computational Biology
• Immunology

Background:

• Pro-inflammatory peptides (PIPs) are crucial mediators in the inflammatory process.
• Accurate identification of PIPs is essential for understanding and managing inflammation.
• Computational methods offer a cost-effective and precise approach for PIP identification.

Purpose of the Study:

• To develop a sophisticated computational predictor, Cat-PIPpred, for identifying pro-inflammatory peptides (PIPs).
• To enhance PIP prediction accuracy by integrating diverse feature sets and advanced machine learning algorithms.
• To provide a reliable tool for functional prediction of specialized peptides.

Main Methods:

• Developed Cat-PIPpred using the CatBoost algorithm combined with cross-modal feature integration.
• Integrated ESM-2 structural embeddings with Dipeptide Deviation from Expected Mean (DDE) evolutionary features.
• Employed feature extraction techniques, refinement protocols, and classifier architecture optimization through cross-validation and independent testing.

Main Results:

• The optimized Cat-PIPpred model demonstrated superior performance compared to existing PIP predictors and general peptide classifiers.
• Feature refinement effectively reduced memory consumption and improved operational efficiency.
• The integration of structural and evolutionary features provided an extensive sequence representation.

Conclusions:

• The study validates the efficacy of integrating multiple feature sets with ensemble learning for accurate peptide prediction.
• Cat-PIPpred offers a reliable and efficient tool for identifying pro-inflammatory peptides.
• The framework provides valuable insights for the functional prediction of specialized peptides in biological research.