Unsupervised AI-platform for Virus-Agnostic Therapeutic Targets Discovery from Multi-Omics Datasets
Year: 2026
Type: article
Abstract: Here, we developed ViralTensors, an unsupervised artificial intelligence (AI) platform that employs Nonnegative Matrix Factorization (NMF) and Weighted NMF (WNMF) for integration of multidimensional datasets and extraction of interpretable latent features. We demonstrate that ViralTensors extracts biologically relevant host regulatory programs and previously unrecognized associations within complex virus–host interactions from global transcriptomic datasets with virus–host protein interaction data. Systems-level analysis of viruses representing three distinct taxonomic families—Coronaviridae (SARS-CoV-2), Filoviridae (Ebolavirus), and Orthomyxoviridae (Influenza virus)—revealed conserved host pathways and targets associated with diverse cellular functions, including endosomal trafficking, gap junction signaling, and cholesterol biosynthesis. The functional relevance of these predicted host targets was validated in cell-based infection assays using FDA-approved therapeutics to inhibit replication of human coronavirus 229E (HCoV-229E), vaccinia virus (VACV-WR), and Hantaan virus (HTNV), representing viruses with RNA and DNA genomes. Furthermore, quantitative global proteomic analysis of Seoul hantavirus infected human endothelial cells independently supported the relevance of AI-predicted host protein networks to viral replication. Collectively, these findings establish ViralTensors as an analytical framework for identifying conserved host dependencies across taxonomically diverse viruses and demonstrate the potential of targeting shared host cellular programs for the development of broad-spectrum antiviral therapeutics.
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7224088635