Deep learning coupled with scalable domain-specific structural validation expands RNA virus discovery from metatranscriptomes

What To Know

The paper presents a two-stage bioinformatics tool: first a pretrained protein language model flags candidate viral sequences, then predicted protein structures (via a sliding-window approach) are checked against a curated database of RNA-dependent RNA polymerase (RdRp) domains to confirm viral identity.

This domain-focused strategy helps detect RdRp domains inside long polyproteins and truncated fragments that can be missed by other methods. Rider was benchmarked across thousands of environmental and clinical metatranscriptomes and reportedly recalls >99% of previously cataloged RNA viruses while identifying additional deeply divergent sequences and clades.

The authors specifically note that in a human IBD cohort Rider detected divergent viral lineages in agreement with other methods and extended detection beyond them.

The work is a computational discovery/annotation advance rather than a clinical trial or treatment study; it provides a scalable pipeline for expanding known RNA viromes and for research into viral associations with human diseases such as IBD.