Dive into machine learning algorithms for influenza virus host prediction with hemagglutinin sequences

Area of Science:

• Virology
• Bioinformatics
• Machine Learning

Background:

• Influenza viruses rapidly mutate, posing public health risks, particularly to vulnerable populations.
• Influenza A viruses have historically caused interspecies pandemics, necessitating accurate origin identification for outbreak prevention.
• Machine learning (ML) is increasingly explored for rapid and precise viral sequence analysis.

Purpose of the Study:

• To evaluate the effectiveness of various machine learning algorithms for predicting the origin of influenza virus sequences.
• To assess ML algorithm performance at different taxonomic levels using real-world data.
• To identify the optimal ML model for rapid and accurate viral sequence origin prediction.

Main Methods:

• Utilized hemagglutinin (HA) sequences, a key protein in the immune response.
• Represented HA sequences using position-specific scoring matrices and word embedding techniques.
• Evaluated multiple machine learning algorithms on real testing datasets with diverse metrics.

Main Results:

• The 5-grams-transformer neural network demonstrated superior performance in predicting viral sequence origins.
• Achieved high accuracy metrics: 99.54% AUCPR, 98.01% F1 score, and 96.60% MCC at higher classification levels.
• Attained strong results at lower classification levels: 94.74% AUCPR, 87.41% F1 score, and 80.79% MCC.

Conclusions:

• The 5-grams-transformer neural network is highly effective for predicting influenza virus origins.
• Accurate viral origin prediction using ML can aid in preventing the spread of outbreaks.
• This approach offers a promising tool for rapid and reliable viral sequence analysis in public health.