Ensemble Learning: Predicting Human Pathogenicity of Hematophagous Arthropod Vector-Borne Viruses

Abstract

AbstractHematophagous arthropods occupy a pivotal role in ecosystems, serving as vectors for a wide array of pathogens with significant implications for public health. Their capacity to harbor and transmit viruses through biting actions creates a substantial risk of zoonotic spillover. Despite the advancements in metagenomic approaches for virus discovery in vectors, the isolation and cultivation of viruses still pose significant challenges, thereby limiting comprehensive assessments of their pathogenicity. Here, we curated two datasets: one with 294 viruses, characterized by 37 epidemiological features, encompassing virus information and host associations; the second with 71,622 sequences of hematophagous arthropod vector-borne viruses, annotated with 33 sequence features. Two XGBoost models were developed to predict arbovirus human pathogenicity—one integrating macroscopic eco-epidemiological data, the other incorporating virus-related sequence features. The macroscopic model identified non-vector host transmission as a key determinant, especially involving Perissodactyla, Artiodactyla, and Carnivora Order. The sequence-based model demonstrated that viral adhesion and viral invasion had distinct trends with consistent increase and decrease in the likelihood of virus pathogenicity to humans, respectively. With validated through an independent dataset, the model exhibited a congruous alignment with documented pathogenicity outcomes. Together, the models offer a holistic framework for assessing the pathogenic potential of viruses transmitted by hematophagous arthropods.