O-GlcNAcPRED-II: an integrated classification algorithm for identifying O-GlcNAcylation sites based on fuzzy...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• Protein O-GlcNAcylation (O-GlcNAc) is a critical post-translational modification impacting cellular processes.
• Aberrant O-GlcNAcylation is implicated in diseases like cancer and neurodegeneration.
• Existing computational methods for identifying O-GlcNAcylation sites lack sufficient prediction sensitivity.

Purpose of the Study:

• To develop an accurate and effective automated computational method for identifying protein O-GlcNAcylation sites.
• To improve upon the sensitivity and overall performance of existing prediction tools.

Main Methods:

• Developed an ensemble model, O-GlcNAcPRED-II, integrating K-means principal component analysis oversampling (KPCA) and fuzzy undersampling (FUS).
• Employed a rotation forest classifier integrating random forest, k-nearest neighbour, naive Bayesian, and support vector machine classifiers.
• Utilized feature space partitioning with four distinct sub-classifiers.

Main Results:

• O-GlcNAcPRED-II achieved high performance metrics: 81.05% sensitivity, 95.91% specificity, 91.43% accuracy, and a Matthew's correlation coefficient of 0.7928.
• The model demonstrated superior performance compared to five existing prediction tools on independent datasets.
• The ensemble approach effectively balanced positive and negative training samples.

Conclusions:

• O-GlcNAcPRED-II represents a significant advancement in the computational prediction of protein O-GlcNAcylation sites.
• The developed method offers improved accuracy and sensitivity, aiding research into O-GlcNAc-related diseases.
• The ensemble strategy and data balancing techniques are effective for enhancing predictive model performance.