Hiding in plain sight: integrating AI with targeted bench methods to discover and characterize viruses in the human body

/ABSTRACT The broad goal of the proposed work is to provide innovative computational tools and laboratory protocols to identify and characterize viruses that are currently unaccounted for in human virome studies. Viruses are abundant in the human body and impact the health and function of human cells and commensal microbial communities, yet understanding of the structure and function of human viromes remains limited. Human viromes appear to be dominated by tailed viruses (phages), yet imperfect computational and experimental approaches limit the ability to broadly recover and recognize all viruses present in an environment. Thus, statements about the human virome and its impacts are currently based only on the fraction of viruses most similar to existing references. The central hypothesis of the proposed work is that the phylogenetic heterogeneity of viruses in the human virome is greater than currently recognized, and that viruses are being missed in two major ways: first, when they are present in sequence datasets and go unrecognized, as “dark matter”; and second, when they are underrepresented in sequence datasets because they are operationally excluded. This hypothesis is addressed through two complementary aims, which develop innovative computational tools and laboratory protocols, and apply these to new and existing sequence datasets in light of metaproteomics, to facilitate and achieve systematic discovery and annotation of the human virome. In Aim 1, computational protein language model (PLM)-based approaches are used to address the technological challenge of identifying and meaningfully characterizing viral sequences in metagenomic data. Novel tools based on PLMs are built to enable viral discovery by the scientific community in any microbiome of interest, and novel conceptual frameworks for genome sequence topology-based classification of uncharacterized viral sequences are developed. In Aim 2, laboratory methods are developed that leverage the distinctive buoyant density features of tailed viruses to systematically deplete these from viromes, thereby allowing detection of viruses otherwise missed. To enable identification of even the most divergent “dark matter” viruses, fractionated viromes are complemented with metaproteomics to reveal as viral those contigs physically co-occurring with the proteins their genes encode (a diagnostic feature of a virion). In both aims, newly developed approaches are applied to stool samples from donors with colorectal cancer (CRC), a disease that is currently increasing in prevalence and in which microbes and viruses have been implicated. Completion of these aims will provide a uniquely broad view of the phylogenetic heterogeneity of the human virome, and new bioinformatic tools and laboratory protocols that will allow the research community to overcome obstacles to viral discovery and characterization currently limiting the translational potential of virome studies. Project Number: 1U01DE035632-01 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of Dental and Craniofacial Research (NIDCR) | Principal Investigator: Kathryn Kauffman (+1 co-PI) | Institution: STATE UNIVERSITY OF NEW YORK AT BUFFALO, AMHERST, NY | Award Amount: $423,366 | Activity Code: U01 | Study Section: Special Emphasis Panel[ZRG1 IIDA-B (50)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11293226