Identifying Pupylation Proteins and Sites by Incorporating Multiple Methods

Area of Science:

• Biochemistry
• Computational Biology
• Microbiology

Background:

• Pupylation is a vital post-translational modification in microorganisms, influencing essential cell functions.
• Accurate identification of pupylation proteins and sites is critical for biological research and drug discovery.
• Current prediction methods require enhancement for improved efficiency and cost-effectiveness.

Purpose of the Study:

• To develop and validate robust computational models for predicting pupylation proteins and specific pupylation sites.
• To enhance the accuracy and efficiency of pupylation prediction, thereby supporting basic biological research and pharmaceutical development.
• To provide researchers with a user-friendly tool for pupylation analysis.

Main Methods:

• Pupylation protein prediction: Feature extraction using KNN, GO annotation, and Word Embedding; dataset balancing with RUS and SMOTE; classification with XGBoost.
• Pupylation site prediction: Utilized six feature extraction methods (TPC, AAI, One-hot, PseAAC, CKSAAP, Word Embedding) and chi-square for feature selection.
• Performance evaluation: Employed rigorous 10-fold cross-validation for both prediction models.

Main Results:

• The pupylation protein prediction model achieved high performance with 95.23% accuracy, 0.8100 MCC, and 0.9864 AUC.
• The pupylation site prediction model demonstrated very high accuracies, outperforming existing methods.
• The developed models offer significant improvements in predicting pupylation events in microbial proteins.

Conclusions:

• The developed computational models provide accurate and efficient tools for identifying pupylation proteins and sites.
• These models can significantly aid in understanding microbial post-translational modifications and accelerate drug discovery efforts.
• The PUP-PS-Fuse web server is now available to facilitate research in pupylation.