MetaFluAD: meta-learning for predicting antigenic distances among influenza viruses

Area of Science:

• Virology
• Computational Biology
• Immunology

Background:

• Influenza viruses constantly evolve, necessitating continuous monitoring of antigenic differences to maintain vaccine efficacy.
• Traditional methods for assessing antigenic differences are time-consuming and labor-intensive, creating a need for efficient computational solutions.
• Understanding viral evolution is crucial for public health and vaccine development.

Purpose of the Study:

• To introduce MetaFluAD, a meta-learning-based computational method for predicting quantitative antigenic distances among influenza strains.
• To model antigenic relationships using hemagglutinin (HA) sequences within a weighted attributed network.
• To enable efficient and accurate prediction of antigenic distances for improved vaccine strategies.

Main Methods:

• Utilized a graph neural network (GNN)-based encoder combined with a meta-learning framework.
• Developed MetaFluAD to learn comprehensive strain representations integrating antigenic and genetic features.
• Applied meta-learning for knowledge transfer across different influenza subtypes, enabling performance with limited data.

Main Results:

• MetaFluAD demonstrated excellent performance and robustness across multiple influenza subtypes (A/H3N2, A/H1N1, A/H5N1, B/Victoria, B/Yamagata).
• The method effectively predicted quantitative antigenic distances, synthesizing GNN encoding and meta-learning strengths.
• Identified dominant antigenic clusters within seasonal influenza viruses, supporting vaccine development and monitoring.

Conclusions:

• MetaFluAD offers a promising computational approach for accurate antigenic distance prediction in influenza.
• The method's ability to transfer knowledge across subtypes makes it valuable for data-limited scenarios.
• MetaFluAD aids in tracking viral evolution and optimizing influenza vaccine design and deployment.