International Edition
Latest News
Health

VHSeek Improves Virus-Host Association Prediction by up to 27%

VHSeek surpasses the iPHoP baseline, boosting genus-level maximum F1 score by 12%, Top-1 accuracy by 19%, and AUPR by 27%, according to a study published in Nature. Predicting virus-host associations with new computational tools improves genus-level maximum F1…

VHSeek surpasses the iPHoP baseline, boosting genus-level maximum F1 score by 12%, Top-1 accuracy by 19%, and AUPR by 27%, according to a study published in Nature.

Predicting virus-host associations with new computational tools improves genus-level maximum F1 score by 12% compared to existing methods, according to a study published in Nature.

Understanding these viral links underpins both disease control and phage therapy. Many current identification tools demand candidate-host genome sequences. These genetic sequences often remain unavailable for various eukaryotic hosts or prove impractical to search during large virome studies.

Researchers built a supervised association predictor named VHSeek to bridge this identification gap. The system integrates DNA-derived and protein-derived viral representations to map connections between viruses and their hosts.

VHSeek combines nucleotide representations from AIDO.DNA with protein representations from ESM2. This dual approach helps the model learn virus-host association patterns before transferring predictions through known host taxonomies from lower to higher ranks.

During inference, VHSeek relies exclusively on viral sequence data and known host taxonomies without requiring host sequence data. When evaluated against iPHoP, researchers found distinct performance gains across multiple evaluation metrics.

Nature reported that VHSeek surpasses the state-of-the-art iPHoP baseline utilized in this research, boosting genus-level maximum F1 score (Fmax), Top-1 accuracy, and area under the precision–recall curve (AUPR) by 12%, 19%, and 27%, respectively.

Confidence thresholding within the architecture enables abstention to reduce false discoveries. By applying a validation-calibrated confidence threshold, VHSeek increases effective host coverage from 36.5% to 64.5% within the processed IMG/VR v4.1 metagenomic collection.

The prediction tool prioritizes candidate viruses and reservoir hosts carrying potential human health impacts. This capability includes identifying multi-host hypotheses where a single virus associates with humans and additional hosts.

One notable example involves the Saboya virus associating with both Homo sapiens and Culicidae, a link not currently captured in standard datasets. These multi-host insights help researchers track viral reservoirs more effectively across diverse environmental samples.

About the author: Health Desk

Health Desk at Archynewsy.