fluspred

fluspred predicts whether Influenza A virus strains are associated with humans by applying sequence-based machine learning models to protein and whole-genome data for zoonotic risk assessment.

Key Features:

• Machine learning models: fluspred applies machine learning to analyze Influenza A protein and genome sequences to predict human association.
• Composition-based feature models: Composition-based features, including dipeptide compositions, are used to develop models for 15 types of influenza A proteins.
• One-hot-encoding protein models: One-hot encoding of protein sequences is employed in protein-based models that achieved an AUC up to 0.99 on validation datasets.
• Genome sequence models: Predictive models built from whole-genome sequences achieved an AUC of 0.98 on validation datasets.
• Performance comparison: The models outperform similarity-based approaches such as BLAST on the same datasets.

Scientific Applications:

• Host association prediction: Identification of Influenza A strains associated with humans to inform surveillance and monitoring.
• Zoonotic potential assessment: Prediction of zoonotic potential and cross-species transmission risk of Influenza A strains.
• Outbreak and pandemic risk support: Support for early detection and monitoring to help prevent outbreaks and assess pandemic risk.
• Host tropism research: Investigation of molecular determinants of host tropism across 15 influenza A proteins and genome-level features.

Methodology:

Machine learning models were trained on Influenza A protein and whole-genome sequences using composition-based features (including dipeptide compositions) and one-hot encoding; models were evaluated by AUC on validation datasets and compared to BLAST.