Using an initial panel of 36 small molecules, the investigators trained a machine learning model to spot certain shapes and chemical features that help stop viruses, scoring each compound’s likelihood of blocking EV71. The authors put their AI-chosen shortlist to the test: out of eight compounds, five successfully slowed the virus in cell experiments about ten times more hits than traditional screening methods usually deliver.

EV71 infections can escalate from mild rash and fever to severe neurological complications, particularly in children under seven and immunocompromised adults. No FDA-approved antivirals currently target the virus. All five confirmed results were tested using computer simulations, which showed that they stuck to certain spots on the virus, findings which could help future researchers stop the virus from changing shape and entering cells. 

We see this as a template for rapid antiviral discovery, added postdoctoral researcher Angela Cesaro, PhD, a study co-author. Whether the next threat is another enterovirus, an emergent respiratory pathogen or a reemerging virus like polio, our AI-driven method shows that, even with limited data, machine learning can accelerate the development of effective solutions and drive a swift response to future outbreaks.

Source: https://www.pennmedicine.org/news/news-releases/2025/april/new-ai-technique-uncovers-antiviral-compounds