Answers are grounded in this paper and cite their sources.
Viral metagenomics and metatranscriptomics have expanded our understanding of viral diversity and associations between viruses and hosts. For eukaryotic viruses, identifying host-related properties remains challenging because infection depends on receptor compatibility, tissue specificity and mechanisms of immune evasion. Interspecies transmission and zoonotic spillover pose major risks to global health, highlighting the need for computational methods that can prioritise plausible host associations and related risks for known and novel viruses. Following PRISMA 2020 guidelines, we systematically evaluated 18 software tools for predicting host-related properties of eukaryote-infecting viruses. The tools were classified into six prediction-target categories, ranging from host-taxon assignment to spillover-risk ranking. We also examined their input requirements, feature representations, modelling strategies and validation approaches. Citation analysis identified 735 tool-citation links, but only 28 documented applications in 27 publications across six tools. Functional testing showed that 16 of the 18 tools returned a host-related prediction for a user-provided sequence. Published performance could not be pooled or directly ranked because tools within the same prediction-target category differed in prediction unit, label space, construction of the held-out evaluation set and reported metric. Tool choice therefore follows from the biological question and the available input, and predicted associations remain hypotheses requiring independent biological validation.
No takes yet. Share an insight, caveat, or question.
Galeeva et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: