Opportunities for machine learning to predict cross-neutralization in FMDV serotype O

Area of Science:

• Veterinary Virology
• Immunology
• Bioinformatics

Background:

• Accurate estimation of cross-neutralization among foot-and-mouth disease virus (FMDV) serotype O strains is essential for effective vaccine selection and disease control.
• Antigenic similarity, measured by r1 values, is a key factor in determining vaccine efficacy against diverse FMDV strains.

Purpose of the Study:

• To develop and validate a machine learning model for predicting FMDV cross-neutralization (r1 values) using VP1 sequence data.
• To identify key viral determinants of antigenic relationships for FMDV serotype O.

Main Methods:

• A dataset of 108 serum-virus pairs from 73 FMDV strains was compiled.
• Machine learning, including Boruta feature selection and random forest classification, was applied to VP1 sequence data and virus neutralization titer (VNT) results.
• Model performance was optimized using tenfold cross-validation and sub-sampling, with predictors including amino acid distances, polymorphisms, and N-glycosylation site differences.

Main Results:

• The machine learning model achieved high accuracy (0.96) in predicting cross-neutralization (r1 values ≥ 0.3) during training.
• The model demonstrated robust performance on independent test sets, with an accuracy of 0.75.
• Specific VP1 residues at positions 48, 100, 135, 150, and 151 were identified as significant predictors of antigenic relationships.

Conclusions:

• Machine learning integrating genomic data offers a powerful approach to estimate FMDV antigenic similarity and predict cross-neutralization.
• This predictive model can significantly aid in guiding vaccine selection and anticipating immune responses to circulating FMDV strains.
• The developed tool is adaptable for other FMDV serotypes and provides a practical means to accelerate vaccine decision-making.