CLCAP: Contrastive learning improves antigenicity prediction for influenza A virus using convolutional neural...

Area of Science:

• Virology
• Computational Biology
• Machine Learning

Background:

• Influenza viruses circulate globally, causing seasonal flu and posing annual public health challenges due to novel variants.
• Rapid viral mutation complicates timely tracking of influenza virus evolution.
• Accurate prediction of antigenic variants is essential for effective vaccine development and preventing viral spread.

Purpose of the Study:

• To develop a fast, low-cost, and precise method for predicting influenza A virus antigenicity.
• To leverage deep learning, specifically a multi-channel convolutional neural network with contrastive learning, for this prediction task.

Main Methods:

• An integrated dataset combining antigenic data and protein sequences was compiled from public resources and literature.
• A multi-channel convolutional neural network architecture was employed, incorporating contrastive learning for enhanced feature representation.
• The model's performance was evaluated on three distinct influenza subtypes (H1N1, H3N2, H5N1).

Main Results:

• The proposed convolutional neural network model significantly outperformed traditional machine learning classifiers in influenza A virus antigenicity prediction.
• The model demonstrated superior performance compared to several state-of-the-art approaches.
• Accuracy improvements were observed at 5.18% for H1N1, 7.03% for H3N2, and 7.82% for H5N1 compared to the best existing methods.

Conclusions:

• The developed framework provides a timely and effective solution for influenza antigenicity prediction.
• The approach shows promise for adaptation to the study of antigenicity in other viral pathogens.
• This method can aid in proactive vaccine strain selection and enhance public health strategies against influenza.