Optimized ensemble learning with multi-feature fusion for enhanced anti-inflammatory peptide prediction

Area of Science:

• Biochemistry
• Computational Biology
• Pharmacology

Background:

• Anti-inflammatory peptides (AIPs) show therapeutic potential but are difficult to discover experimentally.
• Existing computational methods for AIP prediction suffer from data imbalance, poor feature integration, and low accuracy.

Purpose of the Study:

• To develop a more accurate computational model for predicting anti-inflammatory peptides.
• To identify key sequence features that characterize anti-inflammatory peptides.

Main Methods:

• Assembled a high-quality dataset and evaluated feature selection methods (LASSO, PCA, Autoencoder, RF).
• Selected 377 features ranked by Random Forest (RF).
• Implemented and compared deep learning (LSTM, CNN, DNN) and ensemble machine learning classifiers (XGBoost, RF, AdaBoost, GBDT, LightGBM).

Main Results:

• A soft voting strategy combining five ensemble classifiers outperformed existing state-of-the-art predictors.
• Identified significant enrichment of positively charged residues (Lysine, Arginine) in AIPs.
• Observed a predominance of hydrophobic residues in non-AIP sequences.

Conclusions:

• The developed computational approach enhances the prediction of anti-inflammatory peptides.
• Sequence composition analysis provides insights into the molecular basis of AIP activity.
• Findings facilitate the rational design of novel anti-inflammatory peptides for therapeutic applications.