Unisys Researchers Urge AI Drug Models Be Validated Beyond Predictive Accuracy
Unisys researchers call for AI models used in drug repurposing to be evaluated on whether recommendations hold up to biological, safety, and evidentiary scrutiny, not predictive accuracy alone. They warn that high scores can reflect dataset and labeling issues. The team advocates biomedical knowledge graphs and neurosymbolic approaches to build an inspectable evidence chain linking drugs to targets and mechanisms.