Viral Immunogenicity Prediction by Machine Learning Methods.

Area of Science:

• Computational biology
• Immunology
• Virology

Background:

• Viral infections necessitate effective disease control strategies.
• Vaccines are crucial for preventing viral transmission and bolstering immunity.
• Identifying potential vaccine targets computationally is the first step in vaccine development.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting viral protective immunogens.
• To compare the performance of new models against established prediction tools like VaxiJen 2.0.

Main Methods:

• Utilized datasets of 1588 viral immunogens and 468 non-immunogens.
• Employed machine learning algorithms: Random Forest, Multilayer Perceptron, and XGBoost.
• Encoded protein structures using E-descriptors and auto-/cross-covariance methods, selecting relevant features via gain/ratio technique.

Main Results:

• Developed Random Forest, Multilayer Perceptron, and XGBoost models with superior predictive performance on test sets.
• The new models surpassed the predictive accuracy of VaxiJen 2.0.
• Identified hydrophobicity and steric properties as key attributes for viral immunogenicity.

Conclusions:

• Machine learning offers a powerful approach for predicting viral immunogens.
• The developed models represent an advancement over current methods for viral immunogenicity prediction.
• Understanding key protein features can guide future rational vaccine design.