DP-AOP: A novel SVM-based antioxidant proteins identifier.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Antioxidant proteins protect against free radical damage.
• Experimental identification is laborious and costly.
• Existing machine learning models show high accuracy but low sensitivity, suggesting overfitting.

Purpose of the Study:

• To develop an improved machine learning model for accurate antioxidant protein identification.
• To address the low sensitivity issue in existing models.
• To provide a user-friendly tool for researchers.

Main Methods:

• Utilized the SMOTE algorithm for dataset balancing.
• Employed Wei's feature extraction and MRMR for feature selection, reducing dimensions.
• Implemented the Support Vector Machine (SVM) classification algorithm via libsvm.
• Developed a dynamic programming approach for optimal feature subset selection.

Main Results:

• Achieved high performance with 91.076% accuracy, 96.4% sensitivity (SN), 85.8% specificity (SP), 82.6% MCC, and 91.5% F1-score.
• Successfully reduced feature dimensions to 36, then selected 17 optimal features.
• Developed a free web server for public access and further research.

Conclusions:

• The DP-AOP model demonstrates superior performance in antioxidant protein recognition.
• The model effectively balances accuracy and sensitivity, mitigating overfitting.
• The accessible web server will aid future research in antioxidant protein identification.