Recognition for avian influenza virus proteins based on support vector machine and linear discriminant analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Virology

Background:

• Influenza virus hemagglutinin (HA) proteins are crucial for viral entry and are key targets for antiviral strategies.
• Accurate identification of avian influenza virus (AIV) HA proteins is essential for pandemic preparedness and control.

Purpose of the Study:

• To develop and compare computational models for the recognition of AIV HA proteins based on structural characteristics.
• To evaluate the predictive power of Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) models for AIV HA protein identification.

Main Methods:

• Utilized 200 structural properties to represent 400 HA protein sequences from influenza viruses as training samples.
• Developed recognition models using Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA).
• Validated model performance using leave-one-out cross-validation and an external dataset of 200 HA proteins.

Main Results:

• LDA achieved 99.8% accuracy on training samples and 99.5% with cross-validation.
• SVM achieved 99.8% accuracy on training samples and 99.3% with cross-validation.
• External validation showed 95.5% accuracy for LDA and 96.5% for SVM, indicating SVM's superior predictive power.

Conclusions:

• Both SVM and LDA effectively recognize AIV HA proteins based on structural features.
• The SVM model demonstrated superior performance compared to LDA in both internal and external validation.
• These computational models provide a valuable tool for the identification and characterization of AIV HA proteins.