Translating viral genetic data to PRRSV-2 cross-neutralization using machine learning

Area of Science:

• Veterinary Virology
• Computational Biology
• Immunology

Background:

• Porcine reproductive and respiratory syndrome (PRRS) is a major economic threat to the US swine industry, caused by PRRSV-2.
• Current PRRSV-2 control relies on vaccination, but predicting cross-protection between different viral strains remains challenging.
• Virus neutralization (VN) assays are standard but are time-consuming and difficult to scale for rapid decision-making.

Purpose of the Study:

• To develop a machine learning model predicting PRRSV-2 cross-neutralization based on viral genetic sequences.
• To provide a scalable tool for assessing cross-variant vaccine efficacy and guiding swine herd immunization.

Main Methods:

• Generated a dataset of 219 cross-neutralization pairs using PRRSV-2 isolates and antisera.
• Employed machine learning algorithms trained on viral genetic features, including amino acid properties and genetic distances.
• Evaluated model performance using internal and external virus neutralization datasets.

Main Results:

• Machine learning models achieved high prediction accuracy (87-92% balanced accuracy) for PRRSV-2 neutralization.
• Key predictive features included amino acid properties in GP5 and GP2-GP3 epitopes, and overall genetic distance.
• Models demonstrated robust performance across multiple external VN datasets (71-89% balanced accuracy).

Conclusions:

• Developed predictive models for PRRSV-2 cross-neutralization using accessible genetic sequence data.
• A publicly available web tool integrates these models for practical application in veterinary decision-making.
• This approach bridges experimental data and computational methods to enhance PRRSV-2 management strategies.