Tempel

Tempel predicts time-series amino-acid residue mutations in influenza A glycoprotein hemagglutinin sequences to forecast viral evolutionary changes.

Key Features:

• Time-Series Mutation Prediction: Models temporal dynamics of influenza A hemagglutinin amino-acid sequences to predict future mutations.
• Attention-Based Recurrent Neural Networks: Employs recurrent neural networks augmented with attention mechanisms to process sequential residue information.
• Sequential Training Samples: Constructs training samples via strategic splittings and embeddings to capture strain dimensionality over time.
• Residue-Specific Mutation Prediction: Predicts mutations at specific residue sites to provide residue-level forecasts.

Scientific Applications:

• Enhanced Predictive Performance: Experimental results demonstrate improved mutation prediction accuracy compared with traditional approaches across multiple datasets.
• Insights into Viral Evolution: Analyzes mutation dynamics in hemagglutinin to provide interpretable perspectives on influenza A evolution.
• Public Health Preparedness: Informs vaccine strain selection and antiviral strategy by forecasting potential future hemagglutinin mutations.

Methodology:

Constructs sequential training samples from historical hemagglutinin sequence data using strategic splittings and embeddings, and trains attention-augmented recurrent neural networks to capture temporal patterns and predict residue-specific mutations.