AI Metabolic Modeling Identifies Two Drugs That Fight Multiple Viruses

The COVID-19 pandemic exposed a critical gap in global health infrastructure: the absence of ready-made broad-spectrum antivirals capable of containing novel pathogens before vaccines can be developed. This study addresses that gap by developing a computational pipeline that predicts druggable host metabolic targets essential for viral replication across multiple RNA viruses.

The team expanded the human genome-scale metabolic reconstruction Recon 2.2 to include virus-specific biomass reactions representing the molecular building blocks consumed during viral replication. Single-cell RNA sequencing (scRNA-Seq) data from COVID-19 patients were preprocessed using StanDep and fastcore algorithms to generate context-specific metabolic models for individual cell types. Flux balance analysis (FBA) was then applied to predict each cell's capacity to produce virions and to identify enzyme knockouts that selectively impair viral replication without compromising host cell viability.

Applying this workflow to SARS-CoV-2 patient data, the models revealed a striking systemic upregulation of metabolic pathways in upper respiratory tract cells—even in cells not actively infected—suggesting that viral infection reprograms bystander cells to become more permissive to replication. Ciliated and secretory cells showed 2- to 3-fold increases in predicted viral replication capacity. This effect was amplified in patients with severe disease. Expanding the analysis to influenza A and dengue viruses, the team identified a conserved set of metabolic targets enriched for known virus–host interaction partners, validating the biological relevance of the computational predictions.

Two inhibitors were prioritized for experimental validation. Phenformin, a biguanide that inhibits NADH:ubiquinone oxidoreductase (mitochondrial complex I), suppressed SARS-CoV-2 and dengue virus replication in cell culture and demonstrated antiviral efficacy in a Syrian hamster model of SARS-CoV-2 infection in vivo. Atpenin A5, which blocks succinate dehydrogenase (complex II of the electron transport chain), inhibited all four tested viruses—SARS-CoV-2, dengue, respiratory syncytial virus (RSV), and influenza A—with notably high selectivity indices, indicating a favorable therapeutic window. Both compounds target mitochondrial respiratory chain components, reinforcing the concept that viruses broadly co-opt host oxidative metabolism.

These findings establish host cell metabolism—particularly mitochondrial energy pathways—as a tractable and broadly applicable antiviral target space. The modular computational workflow is designed for rapid adaptation to emerging pathogens, offering a scalable tool for pandemic preparedness. Importantly, repurposing existing compounds like phenformin (previously used as an antidiabetic) shortens the path to clinical application.