VirPreNet

VirPreNet predicts influenza A virus virulence from amino-acid sequence data across all eight viral genome segments using ProtVec embeddings and a weighted ensemble convolutional neural network to distinguish virulent and avirulent infections.

Key Features:

• Weighted Ensemble Convolutional Neural Network: Uses a weighted ensemble of convolutional neural networks with a linear layer that integrates segment-specific predictions by assigning different weights for final virulence classification.
• Segment-Specific Analysis: Analyzes all eight influenza A virus genome segments and highlights the contributions of PB2 and HA to virulence prediction.
• ProtVec Embedding: Encodes amino-acid sequences from each segment using ProtVec distributed embeddings as numerical input features.
• Data Labeling and Classification: Classifies infections into avirulent and virulent classes based on mouse lethal dose 50 (LD50) values.
• Validation and Performance: Validated on independent testing data and reported superior, state-of-the-art performance compared to baseline methods.

Scientific Applications:

• Surveillance: Predicts influenza A virus virulence from sequence data to support surveillance of circulating strains.
• Pathogenicity Analysis: Identifies and quantifies segment-specific contributions (notably PB2 and HA) to influenza A virus pathogenicity.
• Research on Determinants of Virulence: Facilitates studies linking sequence variation across the eight segments to virulent versus avirulent phenotypes.

Methodology:

ProtVec embeddings of amino-acid sequences from all eight segments are input to a weighted ensemble convolutional neural network whose segment-specific outputs are integrated via a linear layer assigning weights; labels are derived from mouse LD50 (avirulent vs virulent), and performance is validated on independent testing data.