Bidirectional subsethood of shared marker profiles enables accurate virus classification

Due to the impact of viral metagenomic sequencing, the official virus taxonomy is updated several times a year, with labels being renamed even substantially across releases. While this helps reveal newer aspects on the classification of viruses, existing bioinformatic methods for classification struggle to stay in sync with this ever-improving resource.

We developed a new computer program, named VIRGO, that is able to correctly predict virus families from metagenomic data with an F1 score above 0.9 using a novel viral sequence similarity metric proposed in this work. Moreover, it ensures compatibility with any version of the official taxonomy of viruses.

Virgo is designed to easily incorporate newer releases of the official taxonomy, thus representing a valuable resource in the virology community while raising awareness to develop computational methods that evolve alongside manually curated resources. Video Abstract.