Machine learning prediction of eukaryotic hosts for giant viruses

Giant viruses, classified as Nucleocytoviricota, are known to infect diverse eukaryotes and play crucial ecological roles across various ecosystems. Despite the recovery of tens of thousands of giant virus genomes through cultivation-independent sequencing, the eukaryotic hosts for most of these viruses remain unknown. To address this knowledge gap, researchers developed GVHoP (Giant Virus-Host Predictor), a tool that predicts eukaryotic hosts from giant virus genomes by combining gene content and similarities with eukaryotic sequences.

GVHoP was trained using isolates with experimentally identified hosts and achieved an impressive 97% accuracy in cross-validation. This prediction tool operates at three hierarchical levels of eukaryote classification. Functional analyses revealed that the top predictive features are linked to various processes involved in virus-host interactions, such as viral entry, replication, morphogenesis, and cellular metabolic reprogramming.

When GVHoP was applied to 7,897 giant virus metagenome-assembled genomes, it successfully assigned hosts to 5,280 viruses, uncovering potential novel hosts across major giant virus orders that could not be inferred from core-gene phylogenetic information. The predicted host composition clearly differentiates between aquatic and terrestrial environments, as well as distinct aquatic ecosystems.

GVHoP effectively bridges the gap between giant viruses known only from nucleic acid sequences and their putative hosts across the eukaryotic tree of life, enhancing our understanding of giant virus-host interactions and their potential impacts on ecosystems. This powerful tool paves the way for further ecological and functional studies of environmental giant viruses.