PREDAC-CNN

PREDAC-CNN predicts antigenic variation of influenza viruses by analyzing HA1 amino acid sequences to assess the impact of point mutations on antigenicity.

Key Features:

• Convolutional Neural Network Architecture: Utilizes convolutional neural networks to extract spatial feature representations from HA1 sequences and detect interactions between amino acid sites relevant to antigenic variation.
• Spatially Oriented Representation: Constructs a spatial representation of the HA1 sequence optimized for convolutional frameworks to probe interactions among amino acid sites and their physicochemical attributes.
• Efficient Feature Utilization: Focuses on essential physicochemical properties related to antigenicity and excludes unnecessary amino acid embeddings to streamline prediction and improve accuracy.
• Antigenic Evolution Tracking: Captures intra-HA1 interactions and evaluates the collective effects of point mutations to track antigenic evolution over time.

Scientific Applications:

• Vaccine Strain Recommendation: Predicts antigenic variants and identifies predominant antigenic clusters to inform vaccine strain recommendation, demonstrated on A/H3N2 (1968-2023) and A/H1N1 (1977-2023) datasets.
• Superior Predictive Performance: Validated using 5-fold cross-validation and retrospective testing, showing superior performance compared to existing models.

Methodology:

Convolutional neural networks applied to a spatial HA1 sequence representation optimized for convolution; selection of essential physicochemical property features; validation via 5-fold cross-validation and retrospective testing.