Real-time forecasting of porcine reproductive and respiratory syndrome virus type 2 genetic variant expansion in the field.

Porcine reproductive and respiratory syndrome (PRRS) remains endemic and epidemic in the U.S., driven mainly by the expanding genetic diversity of PRRSV-2. A recently established ORF5-based classification system groups viruses into genetic variants to improve disease tracking. Anticipating which variants are most likely to rapidly increase in incidence, potentially causing epidemic wave-like spread, could significantly improve current monitoring and control efforts. To address this need, we developed a machine learning model to forecast the year-over-year (YoY) growth rate of PRRSV-2 variants, classifying them as fast-growth (>15%) or slow-growth (≤15%) for the next year. The model used 17,158 ORF5 sequences (2015-2024) classified into 191 variants. Thirty features, including exponentially weighted moving averages (EWMAs), genetic distances, and phylogenetic tree metrics, were evaluated. We tested 14 machine learning algorithms and selected the best-performing model, a LightGBM model with 27 features. It achieved 74.1% balanced accuracy with sensitivity of 79.5% for fast-growth variants. The top predictor was the relative difference between a variant's current cumulative sequences and its 3-month EWMA, followed by variant size, other EWMA metrics, and genetic distance features. Emerging variants typically show recent increases in frequency, are represented by at least 50 sequences over the previous three years and exhibit moderate within-variant genetic diversity. The model is retrained quarterly, with updated predictions available through the PRRS-Loom webtool. During the first six quarters of implementation, the model demonstrated improved predictive performance, achieving a balanced accuracy of up to 84% while maintaining a sensitivity of up to 79%. By flagging variants of concern based on fast-growth potential, this approach provides the swine industry with a proactive tool for prioritizing risk management, ultimately improving PRRS control and prevention strategies.