Skip to content

VHSeek model identifies virus families without family genome: collaborative research involving Abu

Share
VHSeek model identifies virus families without family genome: collaborative research involving Abu

Predicting the living organism or family affected by the virus has remained a technical obstacle to epidemic-countermeasure research and antibody-based therapy development, because most available tools require the full genomic sequence of the candidate family, which is not feasible when studying eukaryotic organisms that are not genetically sequenced, or is impractical in the face of the enormous data volume in large environmental virome studies. A new research paper published in the journal Nature Communications revealed a model called VHSeek, a supervised predictor that integrates viral representations derived from nucleic acids and proteins, and is the result of a research collaboration between the University of Southern California and Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi.

The engineering innovation of the model is based oncombining nucleotide representations from the AIDO.DNA model with protein representations from the ESM2 modelto learn the binding patterns between a virus and its family, then transfer the predictions across the known taxonomic classifications of families from lower ranks to higher ranks. The most prominent operational advantage lies in the inference stage, where VHSeek requires only the viral sequence data and the known hierarchical classification of the family, completely freeing itself from the requirement to provide the family genome sequence.

Comparative tests with the iPHoP tool, the most advanced baseline in this field, showed a tangible jump in performance levels, with VHSeek achieving a 12% improvement in the Fmax metric at the genus level, a 19% increase in top-1 accuracy, and a 27% rise in the area under the precision-recall curve (AUPR). The model also employs a reliable probabilistic threshold mechanism thatallows it to consciously refrain from predicting when certainty levels drop, which reduced false statistical discoveries, and enabled it, when applied to the IMG/VR v4.1 metagenomic dataset, to expand effective coverage of families from 36.5% to 64.5% using a confidence threshold calibrated empirically.

The model’s capabilities open a precise horizon for tracking viruses with potential human health impact and identifying their reservoir families, including formulating hypotheses about multiple families in which a single virus is linked to humans and other intermediate families simultaneously, such as observing a link between the Saboya virus and humans and Culicidae mosquitoes, a relationship that had not been documented in previous reference datasets. This research received financial support from the National Science Foundation, the National Institute of Allergy and Infectious Diseases, and the Vetri Fellowship program.

What does this mean for the region

The participation of researchers from Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi, in collaboration with researchers from the University of Southern California, places Gulf academic institutions in a direct contributor role in developing foundational models for microbiology, moving beyond the consumer role of biocomputing technologies toward building the software architecture that drives scientific publication in major globally peer-reviewed journals.

From a field-operations perspective, this model changes the cost and resource equation for public health laboratories and epidemic surveillance centers in the Gulf, Egypt, and the Levant; the ability to detect host families of viruses from environmental samples without the need to survey and sequence the full genome of host organisms shortens laboratory testing budgets, gives regional research teams a rapid early-warning tool for shared diseases, and accelerates the pace of antibody-based therapy research to combat antibiotic-resistant bacteria with the lowest possible computational requirements.

Don't miss the next story

Subscribe for updates