Integrative Modeling of Viral Proteomic Features for Predicting Host Specificity in Plant Viruses: A Genomic-Informed Approach for Epidemiological Monitoring

Plant viruses are the most widespread group of organisms causing plant diseases, representing a major challenge for epidemiological control in our deeply interconnected world, facilitating new encounters between viruses and hosts. Identifying and predicting biological features of viruses may provide a basis for genomic-informed epidemiological monitoring. By combining viral proteome-associated functional features and biological attributes for the well-characterized viral species, our best-performing models reached Cohen’s Kappa scores ranging between 89.8% and 100% in validation for the prediction of horizontal transmission by living vectors. As biological datasets are sensitive to the presence of missing or erroneous labels, our study integrates a pragmatic and innovative methodology that consists of upcycling one-class semi-supervised anomaly detection models to deal with potentially missing labels within the studied biological properties. This allowed us to predict, vectors such as planthoppers, aphids, and whiteflies with Cohen’s Kappa scores of respectively 99.8%, 92,5%, and 92,3%. To further refine our predictive framework, we incorporated an analysis of physicochemical properties and short amino acid kmer frequencies of encoded viral proteins. By leveraging these detailed molecular characteristics, we aimed to enhance the model’s capacity to predict the host plant at the order level with improved accuracy. This approach provides a novel insight into the relationship between viral protein structure and host specificity, offering a potentially valuable tool for preventive identification of susceptible host plants based on viral proteome features.