A novel data augmentation approach for influenza A subtype prediction based on HA proteins

Area of Science:

• Virology
• Bioinformatics
• Computational Biology

Background:

• Influenza A virus poses a significant global health threat, necessitating accurate subtype identification for control and pandemic prevention.
• Genetic diversity, particularly in hemagglutinin proteins, complicates influenza A subtype prediction.
• Current methods face challenges in accurately classifying diverse influenza A subtypes.

Purpose of the Study:

• To introduce PreIS, a novel computational pipeline for precise influenza A subtype classification.
• To leverage advanced protein language models and supervised data augmentation for enhanced prediction accuracy.
• To enable early detection and improve public health preparedness against influenza.

Main Methods:

• Utilized pre-trained protein language models for influenza A hemagglutinin protein sequence analysis.
• Employed supervised data augmentation to expand training datasets without extensive manual annotation.
• Developed a novel pipeline (PreIS) for HxNy subtype classification based solely on hemagglutinin protein sequences.

Main Results:

• PreIS achieved a superior accuracy of 94.54% in influenza A subtype prediction, outperforming the MC-NN model (89.6%).
• Demonstrated proficiency in identifying and classifying unknown influenza A subtypes.
• Established a new benchmark for subtype classification using only hemagglutinin protein data.

Conclusions:

• PreIS offers a highly accurate and efficient method for influenza A subtype prediction.
• The pipeline enhances early detection capabilities, crucial for managing influenza outbreaks and pandemics.
• This research paves the way for improved global influenza surveillance and response strategies.