EnsemGlyPred: Intelligent prediction system for lysine glycation sites integrating deep semantic features and...

Area of Science:

• Biochemistry
• Computational Biology
• Bioinformatics

Background:

• Protein non-enzymatic glycation is implicated in chronic diseases.
• Identifying lysine glycation sites is crucial for understanding disease pathogenesis and developing biomarkers.
• Existing computational methods struggle with complex features and interpretability.

Purpose of the Study:

• To develop an accurate computational method for predicting protein lysine glycation sites.
• To integrate multi-dimensional features and employ weighted ensemble learning for improved prediction accuracy.
• To provide a biologically interpretable and practically useful tool for researchers.

Main Methods:

• Constructed a benchmark dataset from the PLMD database.
• Employed a multi-level feature extraction framework including amino acid composition (AAC), position-specific amino acid composition (PAAC), and deep semantic features (ProGen2).
• Utilized weighted ensemble learning with optimized XGBoost and BiLSTM base classifiers.

Main Results:

• The EnsemGlyPred system demonstrated superior performance over existing methods, especially in recall metrics.
• Ablation experiments validated the effectiveness of multi-feature fusion and weighted ensemble strategies.
• Feature importance analysis identified key amino acids and regions influencing glycation site prediction.

Conclusions:

• EnsemGlyPred offers a robust and interpretable framework for predicting protein glycation sites.
• The study provides a valuable theoretical foundation and practical guidance for computational prediction of post-translational modifications.
• An online platform and datasets are available to facilitate academic research.